API Updates

DeepSeek V4.1 Flash adds native image understanding for coding, document and screenshot workflows. See the model page for current pricing before rollout.

DeepSeek V4.1 Flash

• Model ID: deepseek-v4.1-flash on EvoLink; the page URL is /deepseek-v4-1-flash
• Context window: 1,000,000 tokens, with a maximum output of 384,000 tokens; choose an output budget for the task
• Billing: input, cache hit and output have separate rate entries. Check the model pricing section and final account charges; per-token rates alone do not determine task cost
• Image input: PNG, JPEG, WebP and GIF. Measure representative image usage before scaling
• Protocols: Chat Completions, Messages and Responses take the same model ID; image fields differ: image_url, an image block and input_image
• Thinking and caching: Thinking is enabled by default and prefix caching is automatic; record the thinking setting you send, check the protocol documentation, and read cached tokens from usage
• Legacy routes: on EvoLink, deepseek-v4-flash and deepseek-v4-pro are not affected and continue to serve V4 Flash and V4 Pro; deepseek-v4-flash-vision-exp now redirects to V4.1 Flash. See the migration guide
• View DeepSeek V4.1 Flash | View pricing

GPT Image 2.5 Sunburst and GPT Image 2.5 Flare are now available on EvoLink — the two GPT Image 2.5 models OpenAI released on September 8, 2026, at the same token rates as GPT Image 2.

GPT Image 2.5 Sunburst

• Model ID: gpt-image-2.5-sunburst (no alias; use the exact ID)
• Positioning: OpenAI's pick for workflows where editing precision matters most, with tighter control across multi-turn edits and longer generation times
• Input: text prompt, up to 16 reference images and an optional inpainting mask; output: image
• Quality: five tiers low / medium / high / xhigh / max (default medium); size auto, 15 aspect ratios or explicit WxH; resolution 1K / 2K / 4K
• Billing: token-based — image output, image input, cached input and text input tokens are settled on the upstream usage object; rates in the pricing section
• Endpoint: POST /v1/images/generations, async task with polling or callback_url

GPT Image 2.5 Flare

• Model ID: gpt-image-2.5-flare (no alias; use the exact ID)
• Positioning: OpenAI's fast default for high-quality everyday generation — higher quality than GPT Image 2 at up to 50% lower latency, by OpenAI's figure
• Input: text prompt, up to 16 reference images and an optional inpainting mask; output: image
• Quality: five tiers low / medium / high / xhigh / max (default medium); size auto, 15 aspect ratios or explicit WxH; resolution 1K / 2K / 4K
• Billing: token-based — image output, image input, cached input and text input tokens are settled on the upstream usage object; rates in the pricing section
• Endpoint: POST /v1/images/generations, async task with polling or callback_url

Same token rates do not mean the same bill: each quality tier spends output tokens differently, and GPT Image 2.5 uses a lower tile base per tier than GPT Image 2 (its high costs about what GPT Image 2 medium did). Estimate a request in the Playground or the price calculator before you standardise on a tier.

• Try GPT Image 2.5 Sunburst | Try GPT Image 2.5 Flare | Pricing

The limited-time price drop on Seedance 2.0 Mini and Seedance 2.0 Fast, originally set to end today, is extended through October 6, 2026 (UTC+8). Rates and discount levels are unchanged, and the discount still applies automatically — no changes to your API calls. Seedance 2.0 Standard remains outside the scope of this promotion.

• Seedance 2.0 Mini: 60% off
• Seedance 2.0 Fast: 25% off

The reduction continues to apply across every resolution and every generation mode (text-to-video, image-to-video, reference-to-video). See each model's pricing page for current rates.

The promotion now ends October 6, 2026 (UTC+8), after which both models return to their original pricing.

• Try Seedance 2.0 Mini | Try Seedance 2.0 Fast | Pricing

GPT-6 Astra — OpenAI's most capable model, built for the hardest end-to-end reasoning, coding, computer-use, and research work — is now live on EvoLink at 10% below OpenAI list pricing.

GPT-6 Astra

• Model ID: gpt-6-astra — there is no generic gpt-6 alias
• Context window: 1,050,000 tokens, up to 128K max output
• Billing: uncached input, cache read, cache write and output are metered separately, each 10% below OpenAI list. Current rates are on the GPT-6 Astra model page
• Long context: requests above 272K prompt tokens bill the entire request at long-context rates — 2x for input and both cache roles, 1.5x for output
• Protocols: OpenAI-compatible Chat Completions and Responses, on the same EvoLink API key; OpenAI requires Responses for tool calling
• Reasoning: set reasoning_effort (Chat) or reasoning.effort (Responses) to low, medium, high, xhigh, or max — none and minimal are rejected, and temperature / top_p are not accepted
• Also supported: prompt caching and structured outputs on both surfaces
• Try GPT-6 Astra | View pricing

MiniMax H3 Max — the fast generation variant in MiniMax's Hailuo family — is now live on EvoLink as two routes, with 768p / 480p output and per-second billing.

MiniMax H3 Max

• Model IDs: minimax-h3-max-text-to-video, minimax-h3-max-image-to-video
• Output: 768p or 480p MP4, 5–15 seconds. Aspect ratio 21:9 / 16:9 / 4:3 / 1:1 / 3:4 / 9:16 — text-to-video needs an explicit ratio (defaults to 16:9)
• Pricing: $0.080 per output second at 768p, $0.050 at 480p, same rate on both routes; minimum charge equals the cheapest valid request (480p × 5s, $0.25 per task)
• Image-to-video: send a first frame, a last frame, or both. This route does not support choosing an aspect ratio — the output always follows the input image proportions
• Default resolution: requests that omit quality generate at 768p. Send "quality": "480p" for the cheapest tier
• Not on H3 Max: 2K output, 4-second clips, and reference-driven generation. Use MiniMax H3 when you need 2K, or reference images, video and audio
• Async tasks: submit, then poll status or use callback_url
• Try MiniMax H3 Max | Compare the Hailuo family

Gemini 3.8 Flash is now available on EvoLink, priced the same as Gemini 3.7 Flash.

It is Google's most capable Flash model to date, with clear gains over 3.7 in software engineering, computer use, and multi-step agentic work while keeping Flash-tier throughput and cost.

Gemini 3.8 Flash

• Model ID: gemini-3.8-flash
• Context: 1,048,576 tokens; max output: 65,536 tokens
• Input: text, image, audio, video, PDF; output: text
• Capabilities: thinking, system instructions, structured output, context caching, function calling, code execution, Google Search / Maps grounding
• Pricing: input $0.675/M tokens, output $3.375/M tokens, cache hit $0.0676/M tokens (promotional rate, through December 31, 2026)
• Endpoints: both /v1/chat/completions (OpenAI-style) and the native Gemini API

Thinking levels are unchanged from 3.7

Gemini 3.8 Flash does not support the minimal level, same as 3.7. Available levels are low / medium / high, defaulting to medium.

When migrating from 3.7, if your code sets reasoning_effort: "none" or "minimal" (thinkingLevel: "MINIMAL" on the native API), EvoLink automatically downgrades it to low — your requests will not fail. For explicit control, set low directly.

Beyond the model ID, nothing else needs to change: 3.8 matches 3.7 field for field on context window, max output, modalities, and official pricing. If you are coming from 3.6 Flash or earlier, Google's migration checklist still requires removing custom temperature, top_p, top_k, and candidate_count, replacing the numeric thinking_budget with the thinking_level string, and removing prefilled model turns.

• Gemini 3.8 Flash on EvoLink | Pricing | Models

Claude Fable 5.1 — Anthropic's most capable model, the Fable tier above Opus — is now live on EvoLink at 5% below official token pricing.

Availability

Claude Fable 5.1 can be called on the EvoLink production API with the same EvoLink API key. The Claude Fable 5.1 model page is the source of truth for live pricing and parameters.

Model ID

• claude-fable-5-1 — 1,000,000-token context window, up to 128K max output

Billing dimensions

Five dimensions are metered separately: input, cache write, cache read, output, and the web search tool. Token rates run 5% below Anthropic's official pricing; the web search tool is billed per search at the official rate with no discount, and search results additionally count as ordinary input tokens. See the pricing section for current rates.

Coming from Claude Fable 5

Anthropic's list rates for input, output and cache write are unchanged. The difference is cache read, which Anthropic cut to a quarter of the Fable 5 rate — cache-heavy agent runs and long coding sessions get noticeably cheaper. On EvoLink the other three rates sit about 5.6% above Fable 5, because Fable 5 runs at 10% off list while 5.1 runs at 5% off.

Protocols

OpenAI-compatible Chat Completions and Anthropic Messages, on the same EvoLink API key. Thinking is always on: send thinking: {type: "adaptive"} or omit the parameter, and control depth with effort instead of sampling parameters.

Try Claude Fable 5.1 | View pricing

Anthropic has made its Claude Sonnet 5 price reduction permanent. EvoLink's launch offer becomes the long-term rate at the same time: what you pay per token does not change, and it will no longer rise when the promotion expires.

What changed

• Official list price is now permanently lower: input $3.00 → $2.00/1M tokens, output $15.00 → $10.00/1M tokens, cache write $3.75 → $2.50/1M tokens, cache read $0.30 → $0.20/1M tokens.
• Your EvoLink rate is unchanged. The limited-time launch offer that was scheduled to end on August 31 is now the standard rate, so no expiry increase is coming.
• The discount on the page reads differently. It was measured against the old $3.00 list price; now that the official price is permanently $2.00, the same EvoLink rate shows as 10% below official instead. The number you are billed is identical.

What did not change

The model ID stays claude-sonnet-5, with the same 1M-token context window and 128K max output, on both OpenAI-compatible Chat Completions and Anthropic Messages. No changes to your API calls are required. The web search tool is still billed per search at the official rate.

Current rates are always shown on the Claude Sonnet 5 model page.

Claude Sonnet 5 | View pricing

Gemini Omni 1.1 Flash joins EvoLink's unified video API with five model IDs, output up to 4K, and natively synchronized audio.

Availability

Gemini Omni 1.1 Flash is available through EvoLink's production API. Use the five model IDs below with one EvoLink API key; the Gemini Omni 1.1 Flash model page is the source of truth for live pricing and parameters.

Five Model IDs

• gemini-omni-1.1-flash-text-to-video — generate from a prompt alone
• gemini-omni-1.1-flash-image-to-video — animate a first frame, or interpolate between a first and last frame
• gemini-omni-1.1-flash-reference-to-video — generate from up to 10 reference images and 3 reference videos
• gemini-omni-1.1-flash-video-edit — rewrite objects, style or lighting in one input video
• gemini-omni-1.1-flash-video-extend — continue one input video with matching motion and camera language

Specifications

• Duration: an integer from 3 to 10 seconds, default 3
• Resolution: 360p, 720p, 1080p, 4k, at 24fps (default 720p; Video Edit and Video Extend default to following the input clip)
• Aspect ratio: 16:9, 9:16, auto (default auto)
• Audio: speech, music and sound effects are generated natively and muxed in, always on and free; audio files are not accepted as input
• Input images (i2v): one image as the first frame, or two images as the first and last frame
• Input video: up to 10 seconds for video-edit, up to 30 seconds for video-extend
• Pricing: token-based. See the pricing section on the model page for current rates
• API workflow: submit an asynchronous task to /v1/videos/generations, then poll the task endpoint for status and the final video URL

Model Highlights

• First-and-last-frame control — send two images to image-to-video; their order assigns the roles
• Mixed reference material — reference-to-video takes reference images and reference videos together
• Video extend counts appended seconds — duration is what gets added, so a 3-second input with duration: 3 returns about 6 seconds
• Video edit keeps the input clip as-is — length, aspect ratio and resolution all carry over; set quality when you want a different resolution

Try Gemini Omni 1.1 Flash | Pricing

GLM-5.3 and GLM-5.3 Flash are now available through the EvoLink unified API. Both can be called from Chat Completions and Anthropic Messages.

Availability

Both models can be called on the EvoLink production API. Use the same EvoLink API key with the model IDs below; live pricing and parameters are shown on the GLM-5.3 and GLM-5.3 Flash model pages.

Two model IDs

• glm-5.3 — text-only flagship, for repository-scale coding agents and long task chains
• glm-5.3-flash — natively multimodal, accepts image, video, and file input, for high-volume classification, extraction, and document understanding

Both provide a 1M-token context window and up to 131,072 output tokens, with tool calling and prompt caching.

Try GLM-5.3 | Try GLM-5.3 Flash | View pricing

Wan 3.0 gets a price cut on every route and every resolution, and a speed-oriented tier — Wan 3.0 Prime — joins the family.

New Wan 3.0 Pricing

Effective August 25, 2026 (UTC+8), all three Wan 3.0 model IDs bill at the rates below — 25% off. The change applies automatically; your API calls do not change.

• 480p: $0.0375 / second
• 720p (default): $0.075 / second
• 1080p: $0.15 / second

New: Wan 3.0 Prime

Wan 3.0 Prime is the speed-oriented tier of the same family, available now through the API, with three model IDs for the same core workflow types:

• wan3.0-prime-text-to-video — generate from a prompt alone
• wan3.0-prime-image-to-video — animate a first frame, optionally towards a last frame
• wan3.0-prime-reference-video — generate from reference images, videos, audio, a document, or a web page

Prime's core capabilities align with standard Wan 3.0, with significantly faster end-to-end generation. Use each route's published API schema as the source of truth and validate the request when switching model IDs; EvoLink does not publish a fixed speed multiplier or latency SLA.

Prime rates, 15% off:

• 480p: $0.0578 / second
• 720p (default): $0.119 / second
• 1080p: $0.238 / second

Try Wan 3.0 | Try Wan 3.0 Prime | Pricing

Wan 3.0 (Tongyi Wanxiang 3.0) is coming to EvoLink's unified video API with three model IDs and an all-in-one reference route.

Availability

Wan 3.0 is currently in closed beta. Public access has not opened yet, and we will announce it in a separate entry. The three model IDs below and the Wan 3.0 model page cover the pricing and parameters.

Three Model IDs

• wan3.0-text-to-video — generate from a prompt alone
• wan3.0-image-to-video — animate a first frame, optionally towards a last frame
• wan3.0-reference-video — generate from reference images, videos, audio, a document, or a web page

Pick the route that matches your input: a prompt alone, a first frame, or reference material. The three input types are not interchangeable across routes.

What's New Compared to Wan 2.7

• Durations up to 30 seconds, or duration: -1 to let the model choose
• A 480p tier alongside 720p and 1080p, so iteration gets cheaper
• Documents and web pages as reference material, not just images and video
• An audio track switch (generate_audio), free to toggle either way
• negative_prompt and prompt_extend are not supported — use Wan 2.7 if you depend on them

Try Wan 3.0 | Pricing

DeepSeek V4 Flash Vision (deepseek-v4-flash-vision-exp) is now live — DeepSeek's native multimodal model, which reads text and images in the same request.

What it does

• Image understanding: screenshots, scanned documents, charts, and UI captures, returned as structured text
• All three protocols accept images: image_url on Chat Completions, an image content block on Messages, and input_image on Responses
• Formats: PNG, JPEG, WebP and GIF, sent as a base64 data URI or a public image URL; multiple images per request are supported
• 1M context and 384K max output, same as V4 Flash

Pricing

Identical to DeepSeek V4 Flash — images are billed as input tokens with no surcharge:

• Input (cache miss): $0.441/1M tokens
• Output: $1.324/1M tokens
• Cached input: $0.015/1M tokens

Note

deepseek-v4-flash-vision-exp is an upstream experimental release. Its model name and availability are subject to change, so a fallback route should be kept configured in production.

The text-only models deepseek-v4-flash and deepseek-v4-pro are unaffected by this release. Neither supports image input: the OpenAI-compatible endpoint returns a 400 error, while the Anthropic and Responses endpoints accept the request but replace the image with a placeholder. Requests containing images must specify deepseek-v4-flash-vision-exp.

• DeepSeek V4 Flash Vision | DeepSeek V4 Flash | Pricing

GPT Image 2 now takes a background parameter, so you can ask for a transparent PNG instead of a flat background.

What's New

• New background parameter on gpt-image-2, with two values: opaque (default) and transparent
• transparent returns an image that keeps its alpha channel
• Available through the API and in the Playground, right below Quality

Notes

• transparent is a Preview feature — results may be unstable
• Pricing is unchanged — background does not affect how a request is billed
• gpt-image-2-beta does not support this parameter
• Try GPT Image 2

Seedance 2.5 now generates 1080p video on all five models.

What's New

• 1080p output on seedance-2.5-text-to-video, seedance-2.5-image-to-video, seedance-2.5-reference-to-video, seedance-2.5-video-edit, and seedance-2.5-video-extend
• Pass quality: "1080p" in the API, or select 1080p in the Playground
• Everything else is unchanged: 4–30 second duration, all aspect ratios, synchronized audio, and MOV output

Limited-Time Pricing

Until September 17, 2026 (UTC+8), 1080p is 28% off. The discount applies automatically — no changes to your API calls.

• During the promotion, 1080p costs 1.8× the 720p per-second rate
• After September 17, it returns to 2.5×
• 480p and 720p rates are unchanged throughout

See the pricing page for the exact per-second rate for your account tier.

Notes

• Seedance 2.5 does not support 4K output
• Reference video, image, and audio inputs are unaffected by this change
• Try Seedance 2.5 | Pricing

Effective August 16, 2026 (UTC+8), pricing for DeepSeek V4 Flash and DeepSeek V4 Pro has been increased. The new rates apply automatically, with no changes to your API calls. The model IDs stay deepseek-v4-flash and deepseek-v4-pro.

New pricing

DeepSeek V4 Flash (deepseek-v4-flash):

• Input (cache miss): $0.441/1M tokens
• Output: $1.324/1M tokens
• Cached input: $0.015/1M tokens

DeepSeek V4 Pro (deepseek-v4-pro):

• Input (cache miss): $1.324/1M tokens
• Output: $3.971/1M tokens
• Cached input: $0.044/1M tokens

Both tiers keep the 1M-token context window and 384K maximum output, and remain available on /v1/chat/completions, /v1/messages, and /v1/responses. All other models and capabilities remain unchanged.

• DeepSeek V4 Flash | DeepSeek V4 Pro | Pricing

Gemini 3.7 Flash is now available on EvoLink, priced the same as Gemini 3.6 Flash.

It is the primary agentic workhorse of the Gemini 3 family, with clear gains over 3.6 in code generation and terminal execution while keeping Flash-tier throughput and cost.

Gemini 3.7 Flash

• Model ID: gemini-3.7-flash
• Context: 1,048,576 tokens; max output: 65,536 tokens
• Input: text, image, audio, video; output: text
• Capabilities: thinking, system instructions, structured output, context caching, function calling, code execution, Google Search / Maps grounding
• Pricing: input $0.675/M tokens, output $3.375/M tokens, cache hit $0.0676/M tokens (promotional rate, through December 30, 2026)
• Endpoints: both /v1/chat/completions (OpenAI-style) and the native Gemini API

Thinking levels differ from 3.6

Gemini 3.7 Flash does not support the minimal level. Available levels are low / medium / high, defaulting to medium.

When migrating from 3.6, if your code sets reasoning_effort: "none" or "minimal" (thinkingLevel: "MINIMAL" on the native API), EvoLink automatically downgrades it to low — your requests will not fail. For explicit control, set low directly.

Beyond the model ID, Google's migration checklist requires removing custom temperature, top_p, top_k, and candidate_count, replacing the numeric thinking_budget with the thinking_level string, and removing prefilled model turns.

• Gemini 3.7 Flash on EvoLink | Pricing | Models

From August 14 to December 30, 2026 (UTC+8), Gemini 3.6 Flash and Gemini 3.7 Flash are available at reduced rates. The discount applies automatically — no changes to your API calls. Gemini 3.5 Flash and Gemini 3.5 Flash-Lite are outside the scope of this change.

• Gemini 3.6 Flash: 50% off
• Gemini 3.7 Flash: 50% off

The reduction applies across every billing dimension (input, output, cache hit, cache write, audio input), and input is priced the same across all modalities. See each model's pricing page for current rates.

The promotion ends December 30, 2026 (UTC+8), after which both models return to their original pricing.

• Gemini 3.7 Flash | Gemini 3.6 Flash | Pricing

Seedream 5.0 Pro Layerize is now available on EvoLink at 20% below BytePlus list pricing. It takes a single flat image and splits it into editable layers.

Seedream 5.0 Pro Layerize

• Model ID: doubao-seedream-5.0-pro-layerize
• Input: exactly one png or jpeg image
• Output: one base image plus up to 16 transparent PNG layers
• Layer metadata: every layer returns its stacking order, bounding box, name and description, so you can rebuild the composition on any canvas
• Quality: auto, 1K, 1.5K, 2K
• Prompt: optional. Omit it and the model detects and splits every major element on its own; or describe the elements in plain language; or pin them down with <bbox> coordinates
• API workflow: submit an asynchronous task, then poll the task endpoint for status and the layer URLs

Every output image is priced by its own pixel count, not by the size of the base image. A 2K poster that splits into eight small text layers is billed for one high-tier image and eight low-tier ones.

The number of output images is decided by the split result and cannot be set by request parameters. If any layer fails the whole request fails — there is no partial success, and a failed request is fully refunded.

• Try Seedream 5.0 Pro Layerize | Pricing

DeepSeek V4 Pro and DeepSeek V4 Flash get lower pricing and Responses API support on EvoLink, effective August 13, 2026 (UTC+8). The new rates apply automatically, with no changes to your API calls.

Keep calling deepseek-v4-pro and deepseek-v4-flash — the model IDs are unchanged. They already resolve to the general-availability releases deepseek-v4-pro-0813 and deepseek-v4-flash-0731.

New pricing

DeepSeek V4 Pro (deepseek-v4-pro):

• Input: $0.442/1M tokens
• Output: $0.883/1M tokens
• Cached input: $0.0045/1M tokens

DeepSeek V4 Flash (deepseek-v4-flash):

• Input: $0.148/1M tokens
• Output: $0.295/1M tokens
• Cached input: $0.0030/1M tokens

Responses API support

Both models now support three API endpoints. Select the endpoint that is compatible with your existing integration:

• /v1/chat/completions (OpenAI style)
• /v1/messages (Anthropic style)
• /v1/responses (OpenAI Responses)

Both tiers keep the 1M-token context window and 384K maximum output.

• DeepSeek V4 Pro | DeepSeek V4 Flash | Pricing

Grok Imagine Image 2.0 (grok-imagine-image-2.0) is now available on EvoLink. One model covers both text-to-image and image editing — the number of input images selects the mode automatically.

Highlights

• One model, two modes: send a prompt alone for generation, or add 1–3 input images to switch to editing. Reference each image in the prompt as <IMAGE_0>, <IMAGE_1>, <IMAGE_2>
• Matrix pricing from $0.03/image: quality (Low/Medium) × resolution (1K/2K), every tier 25% below the official xAI rate. Input images are billed once per request
• Batch output: generate 1–10 images in a single request; the output rate scales by n
• 13 aspect ratios plus auto, from square to 20:9 ultra-wide
• Try Grok Imagine Image 2.0

Grok 4.6 and Grok 4.5 are now available on EvoLink through both Chat Completions and Responses. The request model IDs are grok-4.6 and grok-4.5.

Model

• 500,000-token context window with configurable reasoning effort — Grok 4.6 adds an xhigh tier on top of low, medium, and high
• Text and image input, text output
• Works on /v1/chat/completions and /v1/responses with the OpenAI SDK — change the base URL and model name only
• Cached input is billed at a separate, lower rate
• Reasoning tokens are included in output tokens and billed at the output rate

Pricing

Rates are 15% below xAI list pricing. Input and output are identical across both generations:

• Input: $1.70 / 1M tokens
• Output: $5.10 / 1M tokens
• Cached input: $0.425 / 1M tokens on Grok 4.6, $0.256 / 1M tokens on Grok 4.5

The cached tier differs because xAI lists a higher official cached-input rate for Grok 4.6 ($0.50 vs $0.30 per 1M tokens).

Requests whose prompt reaches 200,000 tokens enter the long-context tier, where input, cached input, and output are all billed at 2× — the same threshold and multiplier xAI applies.

Responses Server-Side Tools

On /v1/responses, both models can run xAI server-side tools at the same rates. Only successful server-side calls are charged — ordinary client function calls and failed attempts carry no tool fee, and tool fees are never multiplied by the long-context tier.

• Web search, X search, Code execution: $0.005 per call
• Attachment search: $0.01 per call
• Collections search: $0.0025 per call
• Try Grok 4.6 | Try Grok 4.5 | Pricing

Effective August 11, 2026 (UTC+8), Grok Imagine pricing has been adjusted. The new rates apply automatically — no changes to your API calls.

Grok Imagine Video 1.5 — now 30% off the official rate (previously 20% off):

• 480p: $0.056/s (was $0.064)
• 720p: $0.098/s (was $0.112)
• 1080p: $0.175/s (was $0.20)
• Input image: $0.007 per image (was $0.008)

Grok Imagine 1.0 Beta — updated pricing:

• 480p: $0.0200/s
• 720p: $0.0280/s

See each model's pricing page for live rates.

• Grok Imagine 1.0 Beta | Grok Imagine Video 1.5 | Pricing

Grok Imagine Video 1.0 is now available on EvoLink at 30% below xAI list pricing.

Grok Imagine Video 1.0

• Model IDs: grok-imagine-video-1.0-t2v (text-to-video), grok-imagine-video-1.0-i2v (image-to-video)
• Duration: 1–15 seconds (default 6)
• Resolution: 480p or 720p
• Aspect ratio: 16:9, 9:16, 4:3, 3:4, 3:2, 2:3, 1:1
• Input images (i2v): one image as the first frame, or two to seven images as references
• Pricing: 480p $0.035/second, 720p $0.049/second, plus $0.0014 per input image
• API workflow: submit an asynchronous task, then poll the task endpoint for status and the final video URL

Migration from the Beta models requires only a model ID change; all other parameters remain the same:

• grok-imagine-text-to-video → grok-imagine-video-1.0-t2v
• grok-imagine-image-to-video → grok-imagine-video-1.0-i2v

Two limits differ from the Beta line: duration is 1–15 seconds (Beta allows 6–30), and the mode parameter is not supported.

• Try Grok Imagine Video 1.0 | Pricing

From August 7 to September 6, 2026 (UTC+8), Seedance 2.0 Mini and Seedance 2.0 Fast are available at reduced rates. The discount applies automatically — no changes to your API calls. Seedance 2.0 Standard is outside the scope of this change.

• Seedance 2.0 Mini: 60% off
• Seedance 2.0 Fast: 25% off

The reduction applies across every resolution and every generation mode (text-to-video, image-to-video, reference-to-video). See each model's pricing page for current rates.

The promotion ends September 6, 2026 (UTC+8), after which both models return to their original pricing.

• Try Seedance 2.0 Mini | Try Seedance 2.0 Fast | Pricing

Seedance 2.5 adds two dedicated models and finer output control.

New Models

• seedance-2.5-video-edit: Edit an uploaded video with a prompt — remove, replace, or restyle elements. Output length and aspect ratio follow the input video (duration locked to auto)
• seedance-2.5-video-extend: Extend an uploaded video forward or backward with continuous motion. 4–30 seconds or auto duration

Improvements

• Auto duration: pass duration: -1 (or pick Auto in the Playground) on any Seedance 2.5 model — the model chooses the best length and you are billed by the actual output
• MOV output: output_format: "mov" (H.264 + yuv444p + PCM) for higher color fidelity in editing workflows, at no extra charge
• Try Seedance 2.5

Seedance 2.5 is now available on EvoLink with 3 video generation models supporting text, image, video, and audio reference inputs.

Standard Models

• seedance-2.5-text-to-video: Text-to-video with optional web search for real-time content. 480p/720p, 4–30 seconds, synchronized audio
• seedance-2.5-image-to-video: 1 image = first-frame generation, 2 images = first + last frame generation. Same resolution and duration range
• seedance-2.5-reference-to-video: Multimodal references — images (0–30), videos (0–10), audio (0–10), up to 50 items combined. Supports new generation, video editing, and video extension

What's New vs Seedance 2.0

• Duration doubled: 4–30 seconds in a single pass, up from 4–15 seconds
• Far more references: 50 multimodal items in one request, up from 15
• Audio-only references: reference-to-video accepts audio without any image or video input
• Try Seedance 2.5

Qwen Image 3.0 is now available on EvoLink — Alibaba's Tongyi Wanxiang image generation & editing 3.0, built for information-dense visuals, precise small text, and multilingual typography.

Qwen Image 3.0

• Model IDs: qwen-image-3.0 (standard) and qwen-image-3.0-pro (Pro)
• Modes: Text-to-image (T2I) and image-to-image editing (I2I, 1-3 reference images)
• Prompt: Up to 4,500 tokens
• Output: 1-6 images per request, PNG
• Size: auto, a quality tier (1K/2K), an aspect ratio (1:1, 16:9, 21:9, …), or custom WIDTHxHEIGHT (total pixels 262,144–4,194,304) — works with any n

Highlights

• 10px small-text rendering with photographic detail (micro-expressions, pores, hair strands)
• 12 languages and multilingual typography — posters, menus, storyboards, and UI mockups in one pass
• Two tiers, one contract — switch by changing model; every other request field stays the same
• Qwen Image 3.0 | API Docs

MiniMax H3 now defaults to 768p output and gets a 5% price cut across every tier and user group.

Pricing — 5% off, effective August 5, 2026

• 768p: $0.076 per output second, was $0.080
• 2K: $0.1235 per output second, was $0.130
• Minimum charge: $0.304 per task, was $0.320 (the cheapest valid request is still 768p x 4s)
• Input image surcharge: $0.038 per image, was $0.040 (the first 5 images of a request remain included)
• The cut applies to all three H3 models (text-, image- and reference-to-video) and all user groups. Group-specific rates keep their usual relative discounts

New default resolution: 768p

• Requests that omit quality now generate at 768p ($0.076 per output second) instead of 2K ($0.1235) — about 38% cheaper
• Action required if you rely on 2K: send "quality": "2k" explicitly. Existing calls that never set quality will get cheaper, lower-resolution output
• Both tiers stay fully available on all three models — this changes the default, not the options
• Try MiniMax H3 | Pricing

Grok Imagine Video 1.5 now supports 1080p. The quality parameter accepts a third tier, and every channel behind the route handles it.

What changed

• quality accepts 1080p alongside 480p (default) and 720p — no other parameter changes
• Model ID: grok-imagine-video-1.5-preview — unchanged, no migration needed
• Duration: still 1–15 seconds at every resolution
• Pricing: billed per second, and the per-second rate goes up with resolution — check the pricing table for the current rate at each tier before switching
• Input images: billing is unchanged, still charged per image from 0 to 7
• Try Grok Imagine Video 1.5 | API Docs

Qwen3.8 Max — Alibaba's flagship Tongyi Qwen model — is now live on EvoLink, with a 1M-token context window, controllable thinking, and prompt caching, available over three API protocols.

Qwen3.8 Max

• Model ID: qwen3.8-max
• Context window: 1,048,576 tokens
• Protocols: OpenAI Chat Completions, OpenAI Responses, and Anthropic Messages, all on the same EvoLink API key
• Pricing: Input $1.765/1M tokens, Output $5.295/1M tokens
• Prompt caching: cache write $2.206/1M tokens (125% of input); a cached read costs $0.177/1M tokens with explicit cache_control (10% of input) or $0.353/1M tokens with implicit caching (20%). The two modes are mutually exclusive within a single request
• Controllable thinking: set reasoning_effort to xhigh, medium, or low — mutually exclusive with thinking_budget
• Built-in tools on Chat: web search $0.0005 per call (default turbo strategy) or $0.0006 with search_strategy=max, PDF parsing $0.0030 per page
• Built-in tools on Responses: web search $0.0006 per call, text-to-image search $0.0036 per call, image-to-image search $0.0071 per call
• Tool billing: tool charges follow actual calls and pages, are added on top of token cost, and appear as separate line items in your billing detail
• Try Qwen3.8 Max | API Docs

Effective August 1, 2026, we are adjusting pricing for the models below. GPT-5.6 Sol is outside the scope of this change.

GPT-5.6 Terra

• Input: $1.800/1M tokens, was $2.250 (official $2.00, was $2.50)
• Output: $10.800/1M tokens, was $13.500 (official $12.00, was $15.00)
• Cache read: $0.181/1M tokens, was $0.225 (official $0.20, was $0.25)
• Cache write: $2.250/1M tokens, was $2.814 (official $2.50, was $3.125)

GPT-5.6 Luna

• Input: $0.181/1M tokens, was $0.900 (official $0.20, was $1.00)
• Output: $1.081/1M tokens, was $5.400 (official $1.20, was $6.00)
• Cache read: $0.020/1M tokens, was $0.092 (official $0.02, was $0.10)
• Cache write: $0.225/1M tokens, was $1.125 (official $0.25, was $1.25)
• Try GPT-5.6 | Pricing

MiniMax H3 — the next-generation video model in MiniMax's Hailuo family — is now live on EvoLink as three routes, with 2K output and per-second billing.

MiniMax H3

• Model IDs: minimax-h3-text-to-video, minimax-h3-image-to-video, minimax-h3-reference-to-video
• Output: 2K MP4, 4–15 seconds. Aspect ratio 21:9 / 16:9 / 4:3 / 1:1 / 3:4 / 9:16 — text-to-video needs an explicit ratio (defaults to 16:9), reference-to-video also accepts adaptive
• Pricing: $0.130 per output second at 2K, same rate on all three routes; minimum charge equals the cheapest valid request (2K × 4s)
• Image-to-video: send a first frame, a last frame, or both. This route does not support choosing an aspect ratio — the output always follows the input image proportions
• Reference-to-video: up to 9 reference images, 3 reference videos (15s total) and 3 audio tracks (15s total); reference audio can be the only reference input
• Reference video billing: reference video seconds are added to the output duration — a 5s clip with a 3s reference video is billed as 8s. Reference images and audio add no billable seconds
• Input image billing: the first 5 input images of a request are included; billing starts from the 6th image at $0.04 per image on top of the per-second cost. The minimum charge applies to the per-second part only
• Async tasks: submit, then poll status or use callback_url
• Try MiniMax H3 | Compare the Hailuo family

Suno V5.5 is now available through EvoLink's music generation API and online generator. The Suno page now selects V5.5 by default, while earlier Suno versions remain available.

What's New

• Model ID: use suno-v5.5-beta in API requests
• Custom duration: set duration from 10 to 360 seconds when custom_mode is true
• Generation modes: supports prompt-to-song, custom lyrics/style/title, and instrumental music
• Output: each generation returns two tracks
• Compatibility: existing Suno V4, V4.5, and V5 model options remain selectable

Try Suno V5.5

Grok Imagine Video 1.5 is now available on EvoLink for turning one source image into a 1–15 second video with directed motion and native synchronized audio.

Grok Imagine Video 1.5

• Model ID: grok-imagine-video-1.5-preview
• Generation mode: image-to-video with exactly one source image
• Output: 1–15 seconds at 480p or 720p
• Audio: native dialogue, sound effects, and ambience generated with the video
• Pricing: 480p output $0.064/second, 720p output $0.112/second, plus $0.008 per input image — 20% below xAI official pricing
• API workflow: submit an asynchronous task, then poll the task endpoint for status and the final video URL
• Try Grok Imagine Video 1.5 | API Docs

Effective July 26, 2026, EvoLink has reduced pricing across the Seedream image generation series.

Seedream 5.0 Pro — 25% below official pricing

• Input image: $0.00225 (0.153 Credits) per image
• 1K output: $0.03375 (2.295 Credits) per image
• 2K output: $0.0675 (4.59 Credits) per image
• Try Seedream 5.0 Pro

Seedream 5.0 Lite — 20% below official pricing

• Image generation: $0.028 (1.904 Credits) per image
• Try Seedream 5.0 Lite

Seedream 4.5 and 4.0 — 25% below official pricing

• Seedream 4.5: $0.03 (2.04 Credits) per image
• Seedream 4.0: $0.0225 (1.53 Credits) per image
• Try Seedream 4.5 | Try Seedream 4.0

The latest prices are shown in the EvoLink console and are applied automatically to API usage.

Claude Opus 5 — Anthropic's newest Opus-tier flagship — is now live on EvoLink, at 5% below official token pricing.

Claude Opus 5

• Model ID: claude-opus-5
• Context window: 1,000,000 tokens, up to 128K max output
• Pricing: Input $4.75/1M tokens, Output $23.75/1M tokens — 5% off Anthropic's official $5.00 / $25.00
• Prompt caching: cache write $5.939/1M tokens, cache read $0.475/1M tokens (official $6.25 / $0.50)
• Web search tool: $0.01 per search, billed at the official rate with no discount; search results are billed as ordinary input tokens
• Protocols: OpenAI-compatible Chat Completions and Anthropic Messages, on the same EvoLink API key
• Try Claude Opus 5 | API Docs

Two new Google Gemini models are now live on EvoLink — Gemini 3.6 Flash and Gemini 3.5 Flash Lite. Both offer a 1M-token context window, multimodal input (text/image/video/audio), and built-in reasoning output.

Gemini 3.6 Flash

• Model ID: gemini-3.6-flash
• Context window: 1,048,576 tokens
• Pricing: Input $1.50/1M tokens, Output $7.50/1M tokens — cheaper output than Gemini 3.5 Flash ($9.00)
• Multimodal input: text, image, video, audio at unified pricing
• Built-in reasoning: thinking output is included in standard output billing, no separate dimension
• Prompt caching: automatic caching reduces the cost of repeated long-context requests
• Try Gemini 3.6 Flash | API Docs

Gemini 3.5 Flash Lite

• Model ID: gemini-3.5-flash-lite
• Context window: 1,048,576 tokens
• Pricing: Input $0.30/1M tokens, Output $2.50/1M tokens — our lowest-cost, fastest Gemini tier
• Multimodal input: text, image, video, audio
• Recommended workloads: high-volume, latency-sensitive, and batch tasks — translation, classification, extraction, and simple agentic flows
• Try Gemini 3.5 Flash Lite | API Docs

Moonshot's Kimi K3 is now available on EvoLink. It is built for long-context reasoning, coding, research, and agent workflows, with native visual understanding and tool calling.

Kimi K3

• Model ID: kimi-k3
• Context window: 1,048,576 tokens
• Reasoning: always-on reasoning for complex coding, long-document analysis, and multi-step tasks
• Multimodal input: text, image, and video understanding
• Tool calling: function and tool use for agent workflows
• APIs: compatible with OpenAI Chat Completions and Anthropic Messages
• Prompt caching: automatic caching helps reduce the cost of repeated long-context requests
• Try Kimi K3

Effective July 16, 2026, pricing has been updated for the following models.

Doubao Seed Audio 1.0

• Doubao Seed Audio 1.0 — 0.204 Credits per second ($0.003)
• Model ID: doubao-seed-audio-1-0
• Billing is based on the actual duration of generated audio.
• View Doubao Seed Audio 1.0

DeepSeek V4 Flash & Pro

Following an official DeepSeek price increase, pricing for DeepSeek V4 Flash (deepseek-v4-flash) and DeepSeek V4 Pro (deepseek-v4-pro) has been increased.

See the DeepSeek V4 Flash and DeepSeek V4 Pro pages for current pricing.

All other models and capabilities remain unchanged.

GPT Image 2 Beta Parameter Limits

• Supported: size="auto", or aspect-ratio sizes such as size="1:1" with resolution="1K"
• Auto mode: size="auto" remains supported; if a resolution value is also sent, it is ignored
• Not supported on Beta: resolution="2K", resolution="4K", or explicit WxH pixel sizes such as 1024x1024
• Recommendation: switch to official gpt-image-2 for 2K, 4K, or explicit pixel control

GPT-4O Image Compatibility Migration

GPT-4O Image has reached model retirement, so its compatibility route has been migrated to GPT Image 2 Beta to keep requests available.

• gpt-4o-image and gpt-4o-image-beta now route to gpt-image-2-beta internally.
• Existing requests can continue using the same model IDs for compatibility.
• Billing and routing now follow the GPT Image 2 Beta configuration.
• If size is not provided, the compatibility route keeps the previous square-image default (1:1).
• The mask / mask_url parameter no longer takes effect on this compatibility route; use image_url / image_urls with the prompt for editing-style workflows.

GPT Image 1.5 Legacy Compatibility Migration

The gpt-image-1.5-beta and gpt-image-1.5-lite compatibility routes have been migrated to GPT Image 2 Beta to keep requests available.

• Existing requests can continue using the same model IDs for compatibility.
• Legacy pixel sizes are converted to GPT Image 2 Beta supported aspect-ratio sizes with 1K resolution.
• The quality level parameter no longer takes effect on this compatibility route and will be ignored.
• Billing and routing now follow the GPT Image 2 Beta configuration.

Seedream 5.0 Pro (doubao-seedream-5.0-pro) is now available on EvoLink for Playground testing and API integration. You can generate or edit images through the standard image task workflow and track results with async task status.

Seedream 5.0 Pro

• Model ID: doubao-seedream-5.0-pro
• Status: Live — available in Playground and API workflows
• Workflows: text-to-image, image-to-image, and image editing with reference images
• Output tiers: 1K and 2K clarity, billed independently
• Aspect ratios: 10 presets — 1:1, 2:3, 3:2, 3:4, 4:3, 4:5, 5:4, 9:16, 16:9, 21:9
• Custom pixels: total pixels within [1024x1024, 2048x2048], aspect ratio within [1:16, 16:1]
• Output formats: JPEG (default) and PNG

Key Capabilities

• Ratio + quality auto-sizing: pick a preset ratio and a 1K/2K tier — the model resolves the exact output pixels for you
• Reference-image workflows: up to 10 input images (≤ 30 MB each) for image-to-image and editing; each input image is billed separately
• Async task model: submit a task, then poll /v1/tasks/{task_id} or use a callback_url
• Image-based billing: credit and image usage are returned for this image-priced route

Documentation

• Seedream 5.0 Pro

Krea 2 Turbo

Krea 2 Turbo is now available — a premium text-to-image generation model optimized for cinematic visuals and poster-grade composition, with selectable 1K and 2K output tiers.

• Dual output tiers: 1K (standard) and 2K (high-definition) — billed independently
• 11 aspect ratios: 1:1, 4:3, 3:4, 5:4, 4:5, 2:3, 3:2, 9:16, 16:9, 1:2, 2:1
• Try Krea 2 Turbo

Nano Banana 2 Lite Beta (nano-banana-2-lite-beta) is now available on EvoLink as an experimental 1K route for evaluation, prototypes, and cost-sensitive non-critical batches. For production workloads that require stronger availability, keep Nano Banana 2 Lite official as the fallback.

Nano Banana 2 Lite Beta

• Model ID: nano-banana-2-lite-beta
• Status: experimental Beta route
• Output: fixed 1K image generation
• Workflows: text-to-image, image-to-image, and image editing with reference images
• Billing: per successful request
• API: use the same image generation endpoint; switch only the model parameter
• Availability: may change during Beta; keep Official Lite as fallback for production workloads
• Note: the Beta version does not support thinking-level controls
• Try Nano Banana 2 Lite Beta

This model update includes two changes: Nano Banana 2 Lite is now live on EvoLink, and Claude Fable 5 is available again after the temporary access restriction noted on June 15.

Nano Banana 2 Lite

Nano Banana 2 Lite is based on Google Gemini 3.1 Flash-Lite Image and is built for fast 1K image generation, image editing, and high-volume creative iteration.

• Model ID: gemini-3.1-flash-lite-image
• Capabilities: text-to-image, image editing, and multi-image composition
• Output: native 1K with 14 aspect ratio options
• Speed: fast generation for rapid iteration
• Try Nano Banana 2 Lite

Claude Fable 5 availability

Claude Fable 5 (claude-fable-5) is live again on EvoLink. The model page and API access are back to the normal available state.

• Model ID: claude-fable-5
• Status: available on EvoLink
• Context window: 1M tokens
• Max output: 128K tokens
• Best fit: the hardest coding tasks, long-horizon agents, and high-stakes long-context work where the premium over Opus 4.8 is justified
• Try Claude Fable 5

Google Gemini Omni Flash is now live on EvoLink — a fast video model that covers text-to-video, image-to-video, reference-to-video, and video editing through a single API key. Unlike generation-only routes, Gemini Omni Flash treats editing as a first-class workflow for short-video iteration.

Gemini Omni Flash

• Model IDs: gemini-omni-flash-text-to-video, gemini-omni-flash-image-to-video, gemini-omni-flash-reference-to-video, gemini-omni-flash-video-edit
• Text to Video: generate a 720p video from a prompt, 3–10s or Auto duration
• Image to Video: animate one input image with a prompt, 3–10s or Auto duration
• Reference to Video: guide the scene with 1–6 reference images, 3–10s or Auto duration
• Video Edit: edit one MP4 input (up to 10s) with a natural-language instruction
• Workflow: async task with callback support — no Google Cloud project required
• Billing: token-based
• Try Gemini Omni Flash

Claude Sonnet 5 from Anthropic is now live on EvoLink — the best balance of speed, intelligence, and cost for coding, agentic workflows, and everyday tasks, with a 1M-token context window and up to 128K max output.

Claude Sonnet 5

• Model ID: claude-sonnet-5
• Context window: 1M tokens
• Max output: 128K tokens
• Capabilities: strong coding and technical analysis, reliable agentic tool use, adaptive thinking (on by default), vision input, web search tool
• Recommended workloads: day-to-day coding, agent-heavy workflows, and large-context tasks where balanced cost matters
• Billing: token-based — Input, Output, Cache Write, Cache Hit; web search tool billed per request
• API: same surface as the Claude Messages API — adaptive thinking and prompt caching supported
• Try Claude Sonnet 5

Effective July 7, 2026, 00:00 (UTC), to continue providing stable and reliable service, we are updating the pricing for two image generation models:

• Seedream 4.5 — 2.375 credits per image (≈ $0.0349)
• Seedream 4.0 — 1.8 credits per image (≈ $0.0265)

Notes

• Billed per generated image; batch requests are charged as the unit price × n.
• The new pricing takes effect automatically at the time above. Requests made before then are billed at current rates.
• All other models are unaffected.

If you have any questions, please contact our support team.

Effective June 29, 2026, to continue providing stable and reliable service, we are updating the pricing for Doubao Seed Audio 1.0:

• Doubao Seed Audio 1.0 — 0.17 credits per second (≈ $0.0025)

Notes

• Billed by audio duration (per second of generated audio).
• All other models and capabilities remain unchanged.

If you have any questions, please contact our support team.

Doubao Seed Audio 1.0 is now available — a multimodal audio generation model that turns text into lifelike speech, clones voices from reference audio, and can even generate audio guided by a reference image.

Capabilities

• Text-to-speech: Hundreds of built-in voices across multiple languages, dialects, and character styles, with adjustable speed, loudness, and pitch
• Voice cloning: Pass reference audio to generate speech in a consistent target voice (up to 3 references per request)
• Image-guided generation: Provide a reference image to produce audio that fits the visual context

Highlights

• Transparent per-second billing — pay only for the audio you generate
• Up to 120 seconds of output, multiple formats (MP3 / WAV / PCM / OGG Opus)
• OpenAI-compatible asynchronous API
• Try Doubao Seed Audio

Seedance 2.0 Mini is now available — a lightweight, lower-cost variant of the Seedance 2.0 family at roughly half the price of the standard models, ideal for rapid iteration and cost-sensitive scenarios.

Model Variants

• seedance-2.0-mini-text-to-video: Text-to-video generation. 480p / 720p, 4–15 seconds, native audio sync
• seedance-2.0-mini-image-to-video: 1 image = first frame, 2 images = first + last frame. Same resolution and duration range
• seedance-2.0-mini-reference-to-video: Multimodal reference input — images, videos, and audio. Supports video editing and extension

Highlights

• Supports AIGC-generated realistic human materials
• Same API and endpoint as the rest of the Seedance 2.0 family — just switch the model parameter
• Try Seedance 2.0 Mini

Topaz Video Upscale — AI Video Super Resolution

Topaz Video Upscale is now available on EvoLink — enhance video resolution with AI-powered upscaling up to 4x. Per-second billing starting at 3.7 credits/s.

• Model ID: topaz-video-upscale
• Input: Video URL (MP4, MOV, MKV — max 50MB)
• Upscale factors: 1x (quality enhancement), 2x (double resolution), 4x (quadruple resolution)
• Processing: Async task — submit and poll or use callback URL
• Pricing: 3.7 credits/s (1x/2x), 5.92 credits/s (4x). Result URLs valid for 24 hours
• Topaz Video Upscale API

Seedance 2.0 — 4K Resolution Output

Seedance 2.0 (Standard) now supports 4K video output across all three modes — text-to-video, image-to-video, and reference-to-video. Billed per second; see the pricing table for current 4K rates.

• Modes: text-to-video, image-to-video, reference-to-video (Standard only; Fast and Mini keep their existing resolution sets)
• Resolutions: 480p, 720p, 1080p, 4K
• Output: MP4 (H.264) + AAC audio, 4–15s, 24fps
• Pricing: per second — check the pricing table for exact 4K rates
• Seedance 2.0 API

HappyHorse 1.1 is now live on EvoLink with three task variants for text-, image-, and reference-driven video generation.

HappyHorse 1.1

• Model IDs: happyhorse-1.1-text-to-video, happyhorse-1.1-image-to-video, happyhorse-1.1-reference-to-video
• Resolution: 720p / 1080p
• Duration: 3–15s for all three variants
• Aspect ratios: 9 ratios (16:9, 9:16, 1:1, 4:3, 3:4, 4:5, 5:4, 9:21, 21:9) for T2V/R2V; I2V inherits the source image dimensions
• Reference-to-Video: up to 9 reference images mapped to character1, character2, … in the prompt
• HappyHorse 1.0 update: T2V/R2V now also support all 9 aspect ratios
• Try HappyHorse 1.1

Midjourney V8.1 is now available — sharper, native HD output with a streamlined set of 7 operations and OpenAI-compatible quality control.

Image Generation

• mj-v8.1: Text-to-image & image-to-image, 4 images per request, native MJ prompt syntax
• Quality (top-level quality): standard or hd — HD is native 2K (1.5× price)
• Speed (model_params.speed): Draft / Fast / Turbo
• Route parameters: --ar, --s, --chaos, --exp, --raw, --seed, --sref, --oref, --iw, --sw, --ow

6 Editing Operations

• Variation (subtle/strong), Remix (regenerate with a new prompt)
• Canvas Edit (reposition on canvas + repaint), Retexture (style/material transfer)
• Advanced Edit (upload image + polygon mask + canvas), Remove Background
• Try Midjourney V8.1

Playground

• Visual controls for speed, quality, canvas layout, and polygon mask areas
• Task ID workflow for Variation / Remix / Canvas Edit
• Real-time price estimate (speed × quality)

API

• OpenAI-compatible top-level quality parameter (standard / hd)
• Per-variant API reference with copy-ready examples on /v1/images/generations

Kling 3.0 Turbo is now live on EvoLink — a faster variant of Kling 3.0 for both text-to-video and image-to-video generation.

Highlights

• Two modes: Kling 3.0 Turbo covers text-to-video and image-to-video
• Output: 720p and 1080p, 3–15 second clips with per-second billing
• Pricing: from 7.2 credits/second at 720p; 1080p billed at 1.25×
• One integration: same API key and endpoint as the rest of the Kling family — switch with a single model parameter (kling-v3-turbo-text-to-video / kling-v3-turbo-image-to-video)

Try it now in the playground.

We are excited to announce the launch of GLM-5.2, Z.ai's latest flagship model, now available on EvoLink.

GLM-5.2

• Model ID: glm-5.2
• Context Window: ~1M tokens, billed at a single flat rate with no long-context surcharge
• Features: Deep thinking (thinking), tool calling / function calling, prompt caching (cache reads)
• Endpoint: OpenAI-compatible /v1/chat/completions

Key Use Cases

• Coding Agents: Connect coding agents, coding CLIs, and editor tools through the OpenAI-compatible endpoint, plus repo Q&A, code generation, and review workflows
• Agentic Tool Use & Function Calling: Plan, call tools, and recover from errors across multi-step agentic workflows
• Long Context: Full codebase analysis and large-document reasoning with ~1M context
• GLM-5.2 API

Due to an upstream access-policy change on Anthropic's side, Claude Fable 5 (claude-fable-5) is temporarily unavailable. We're actively working to restore it and will share a recovery timeline as soon as we have one.

In the meantime

• Alternative: switch to Claude Opus 4.8 — also 1M context, with the same API surface as Fable 5, so migration is just a model-ID swap
• Scope: only claude-fable-5 is affected; the rest of the Claude line (Opus / Sonnet / Haiku) remains fully available
• Updates: we'll post the recovery notice here in the ChangeLog as soon as it's back

We apologize for the inconvenience.

Claude Fable 5 from Anthropic is now live on EvoLink — Anthropic's most powerful and most intelligent model, a new tier above Opus, built for the hardest coding, long-horizon agent, and high-stakes long-context work, with a 1M-token context window.

Claude Fable 5

• Model ID: claude-fable-5
• Context window: 1M tokens
• Max output: 128K tokens
• Capabilities: state-of-the-art reasoning, advanced coding, long-horizon agentic tool orchestration, vision input
• Recommended workloads: the hardest coding tasks, longest agent runs, and highest-stakes long-context decisions where stronger reasoning is worth a premium
• Pricing: $10 / 1M input, $50 / 1M output — about 2× Opus 4.8; web search tool billed per request
• API: same surface as Opus 4.8 — adaptive thinking only (no temperature / top_p / budget_tokens), effort up to max; do not send thinking: disabled (omit the parameter instead)
• Routing: keep Opus 4.8 as the strong default and escalate only frontier-difficulty requests to Fable 5
• Try Claude Fable 5

Due to cost adjustments, pricing for Veo 3.1 Fast (veo3.1-fast-beta) and Veo 3.1 Pro (veo3.1-pro-beta) has been increased, effective immediately.

See the Veo 3.1 Fast and Veo 3.1 Pro pages for current pricing.

In addition, the 1080p and 4K resolution options for these two models have been retired — only 720p is now available.

We are excited to announce the launch of MiniMax-M3, MiniMax's latest multimodal model, now available on EvoLink.

MiniMax-M3

• Model ID: MiniMax-M3
• Context Window: ~1M tokens (base pricing within 512K; beyond 512K billed at 2× as long context)
• Features: Deep thinking (thinking), multimodal input (image / video / PDF), prompt caching
• Dual Endpoints: OpenAI-compatible /v1/chat/completions + native Anthropic Messages /v1/messages (for Claude Code–style CLIs)

Key Use Cases

• Coding Agents & Claude Code: Native /v1/messages endpoint for Claude Code–style CLIs, plus repo Q&A, code generation, and review workflows
• Multimodal: Image, video, and PDF input for visual reasoning and document understanding
• Long Context: Full codebase analysis and large-document reasoning with ~1M context
• MiniMax-M3 API

Claude Opus 4.8 from Anthropic is now live on EvoLink — Anthropic's most capable Opus model for complex reasoning, long-horizon agentic coding, and high-autonomy workflows, with a 1M-token context window.

Claude Opus 4.8

• Model ID: claude-opus-4-8
• Context window: 1M tokens
• Max output: 128K tokens
• Capabilities: hybrid reasoning, advanced coding, agentic tool orchestration, vision input
• Recommended workloads: frontier coding agents, complex reasoning, long-context analysis, high-autonomy automation
• Billing: token-based — Input, Output, Cache Write, Cache Hit; web search tool billed per request
• Migration: drop-in replacement for Opus 4.7 — same API and pricing tier; Opus 4.7 remains available
• Try Claude Opus 4.8

The following models have been retired from EvoLink and are no longer available for new requests. Please migrate any remaining integrations to the recommended alternatives below.

Retired models

• GPT Image 1 — gpt-image-1
• GPT Image 1.5 (Beta) — gpt-image-1.5-beta (alias gpt-image-1.5-lite)
• GPT-4O Image — gpt-4o-image
• Kimi K2 Series — kimi-k2-thinking, kimi-k2-thinking-turbo

Recommended alternatives

• Image generation → GPT Image 2
• For other workloads, see the full model catalog

Wan 2.7, the next-generation video model from Alibaba's Tongyi Wanxiang team, is now live on EvoLink. It brings text-to-video, image-to-video, reference video, and video editing together under a single /v1/videos/generations endpoint — just switch the model parameter to change modes.

Wan 2.7 Text-to-Video

• Model ID: wan2.7-text-to-video
• Features: Generate cinematic video from text prompts, with optional audio input and seed control
• Duration: 2-15 seconds, 720p / 1080p
• Highlights: Prompts up to 5000 characters with negative prompt support

Wan 2.7 Image-to-Video

• Model ID: wan2.7-image-to-video
• Features: Animate images into video with first frame, last frame, or both under your control
• Duration: 2-15 seconds, 720p / 1080p
• Highlights: Precise first/last frame control with driving audio and clip input

Wan 2.7 Reference-Video

• Model ID: wan2.7-reference-video
• Features: Generate clips driven by reference images, videos, and audio
• Highlights: Reference-guided for consistent style and content

Wan 2.7 Video Edit

• Model ID: wan2.7-video-edit
• Features: Edit existing videos with text instructions and up to 4 reference images
• Highlights: No re-rendering required — keep original audio or generate new audio
• Wan 2.7 API

Gemini 3.5 Flash from Google is now live on EvoLink — a next-generation Flash-tier model with multimodal input (text/image/video/audio) at unified pricing and built-in reasoning output.

Gemini 3.5 Flash

• Model ID: gemini-3.5-flash
• Context window: 1.05M tokens
• Multimodal input: text, image, video, audio — audio priced at the same rate as text input
• Built-in reasoning: thinking output included in standard output billing, no separate dimension
• Recommended workloads: multimodal reasoning, audio/video understanding, planning, balanced agentic workflows
• Billing: 4-dimension token billing — Input, Output, Cache Hit, Audio Input
• Try Gemini 3.5 Flash

The entire HappyHorse 1.0 video generation lineup has been reduced in price, effective immediately for all users.

Affected Models

• happyhorse-1.0-text-to-video
• happyhorse-1.0-image-to-video
• happyhorse-1.0-reference-to-video
• happyhorse-1.0-video-edit

gpt-image-2-beta now supports the resolution parameter for output image sharpness control.

What's New

• New resolution parameter: 1K / 2K / 4K (case-insensitive, default 1K)
• Combined with size (aspect ratio), resolution controls the output pixel budget — higher tiers produce sharper, larger images
• Size allowlist aligned with gpt-image-2 official, now including 4:5 and 5:4 ratios
• Default size changed to auto (aligned with official behavior)

Documentation

• GPT Image 2 Beta API

GPT Image 1 and GPT Image 1.5 (Official) have migrated from flat per-image pricing to granular token-based billing, aligned with OpenAI's upstream usage reporting.

What Changed

• GPT Image 1 now uses 5-dimension token billing: Image Output, Image Input, Image Cached Input, Text Input, Text Cached Input. Output Text + Thinking is included at no extra charge.
• GPT Image 1.5 Official now uses 6-dimension token billing: same 5 dimensions plus a dedicated Output Text + Thinking dimension (typically less than 1% of total cost).
• GPT Image 1.5 Beta (Lite variant) remains unchanged at flat per-request billing.

Why It Matters

• Pricing now reflects actual resource consumption per request instead of a single flat fee.
• Low-quality / small-size generations cost significantly less; high-quality / large-size generations with many reference images cost more.
• Usage logs now display a full token breakdown by dimension, making cost attribution transparent.

Billing Dimensions at a Glance

• GPT Image 1 Pricing
• GPT Image 1.5 Pricing

EvoLink Moderation 1.0 is now live, offering multimodal content moderation for text and images with 13 harm categories and a deterministic risk-level summary.

EvoLink Moderation 1.0

• Model ID: evolink-moderation-1.0
• Endpoint: OpenAI-compatible /v1/moderations
• Inputs: text, single image, or text + single image (one image per request)
• Categories: 13 harm categories — harassment, hate, illicit, self-harm, sexual, sexual/minors, violence, plus their subtypes
• Output: standard moderation fields plus evolink_summary with risk_level (low/medium/high), violations, max_score, and max_category for one-line decisions
• Thresholds: per-category cutoffs tuned for production — strict on minor-safety and self-harm, relaxed on harassment/hate
• Billing: 1 credit per call — no token-based billing or volume tiers
• Try EvoLink Moderation 1.0

HappyHorse 1.0 is now live on EvoLink with four task variants for text, image, reference, and edit-driven video generation.

HappyHorse 1.0

• Model IDs: happyhorse-1.0-text-to-video, happyhorse-1.0-image-to-video, happyhorse-1.0-reference-to-video, happyhorse-1.0-video-edit
• Resolution: 720p / 1080p
• Duration: 3–15s for T2V/I2V/R2V; video-edit follows the source video (capped at 15s)
• Aspect ratios: 5 ratios (16:9, 9:16, 1:1, 4:3, 3:4) for T2V/R2V; I2V and video-edit inherit source media dimensions
• Reference-to-Video: up to 9 reference images mapped to character1, character2, … in the prompt
• Video Edit: 1 input video + up to 5 reference images, optional preserve-original-audio
• Billing: per-second — output duration for T2V/I2V/R2V; (input + output) duration for video-edit
• Try HappyHorse 1.0

GPT Image 2 (official) is now available on EvoLink with token-based billing and flexible size controls.

GPT Image 2

• Model ID: gpt-image-2
• Size: auto (model decides), 15 aspect ratios (1:1, 16:9, 21:9, 2:3, 1:2, 1:3, …), or custom pixels — multiples of 16, total 655,360–8,294,400 pixels, longest edge ≤ 3840, aspect ratio ≤ 3:1
• Resolution: 1K / 2K / 4K — applied when size is an aspect ratio (e.g., 4K + 16:9 → 3840×2160)
• Quality: low / medium / high — drives output token count
• Batch: 1–10 images per request
• Image-to-image / edit: reference images via image_urls
• Billing: token-based across Image Input, Image Cached Input, Image Output, Text Input, Text Cached Input
• Over-budget auto-fit: combinations beyond the 8.29 MP ceiling are scaled down (e.g., 4K + 1:1 → 2880×2880)
• Try GPT Image 2

GPT Image 2 Beta is now available on EvoLink as an early-access route, ahead of the official GPT Image 2 release.

GPT Image 2 Beta

• Model ID: gpt-image-2-beta
• Supports 14 aspect ratios: auto, 1:1, 3:2, 2:3, 4:3, 3:4, 16:9, 9:16, 21:9, 9:21, 2:1, 1:2, 3:1, 1:3
• Supports image-to-image workflows with reference images
• Flat per-call pricing — one charge per request, independent of output size
• Try GPT Image 2 Beta

Seedance 2.0 model family now supports AIGC-generated realistic human materials across all variants.

• Use AIGC-generated character images, motion videos, and voice audio as input materials
• Available for all Seedance 2.0 models (Standard and Fast)
• Try it on Seedance 2.0

Midjourney V7 is now available with 12 model operations, a visual Prompt Builder, and complete API documentation.

Image Generation

• mj-v7: Text-to-image with native MJ prompt syntax, 4 images per request
• 3 speed tiers: Draft (1x), Fast (2x), Turbo (4x)
• Full parameter support: --ar, --s, --chaos, --w, --exp, --q, --no, --seed, --raw, --tile
• Image references: Image prompts (--iw), Style Reference (--sref/--sw/--sv), Object Reference (--oref/--ow)

11 Editing Operations

• Variation (subtle/strong), Remix (strong/subtle mode), Upscale (standard/creative)
• Enhance (draft to standard quality), Pan (directional extend), Outpaint (expand all directions)
• Inpaint (mask-based region repaint), Canvas Edit (reposition + fill), Retexture (style transfer)
• Remove Background, Advanced Edit (upload + mask + canvas)
• Try Midjourney V7

Visual Prompt Builder

• Build MJ prompts visually with sliders, toggles, and presets — no syntax memorization needed
• Real-time assembled prompt preview with color-coded highlighting (URLs blue, params green)
• Upload image prompts, style references, and object references with drag-and-drop
• Smart dependency warnings: incompatible parameter combinations detected automatically

Playground

• Task ID display: hover on generated images to copy Task ID for secondary operations
• task_id + image_number inputs for all editing operations
• Full parameter controls: type, mode, direction, scale, mask upload, canvas dimensions

Model Pricing Adjustments

Due to upstream pricing changes from our providers, the costs of certain models have been updated. To ensure the stability and sustainability of our platform, we will adjust the pricing of affected models accordingly.

• Affected models: Google series and Claude series
• Effective date: April 8, 2026, 00:00 Pacific Time
• Updated pricing: Please refer to the Models page

Credit Exchange Rate Adjustment

Due to recent fluctuations in the USD exchange rate, the credit exchange rate for top-ups has been adjusted to more accurately reflect current market conditions and maintain platform sustainability.

• Effective date: April 7, 2026
• Updated rates: Please refer to the latest rates displayed on the top-up page

We will continue to monitor upstream pricing dynamics and exchange rate changes, and strive to provide a stable, reliable, and seamless gateway service. Thank you for your understanding and support.

Seedance 2.0 is now available with 6 video generation models supporting text, image, video, and audio reference inputs.

Standard Models

• seedance-2.0-text-to-video: Text-to-video with optional web search for real-time content. 480p/720p, 4–15 seconds, native audio sync
• seedance-2.0-image-to-video: 1 image = first-frame generation, 2 images = first + last frame generation. Same resolution and duration range
• seedance-2.0-reference-to-video: Multimodal references — images (0–9), videos (0–3), audio (0–3). Supports new generation, video editing, and video extension

Fast Models

• seedance-2.0-fast-text-to-video, seedance-2.0-fast-image-to-video, seedance-2.0-fast-reference-to-video
• Faster processing with the same input capabilities as the standard models
• Try Seedance 2.0

Suno Persona (suno-persona) is now available — create reusable music personas from completed Suno music tracks, then apply them to future generations for consistent vocal and style characteristics.

What's New

• Suno Persona Creation: Extract vocal and style features from any completed Suno track to create a reusable persona
• persona_id: Use the returned persona_id in any Suno music generation request via the persona_id parameter
• persona_model: Choose between style_persona (all models) and voice_persona (V5 only)
• Pricing: 1 credit per persona creation
• Try Suno Persona

Music Generation Updates

• New parameters: persona_id and persona_model added to Suno music generation API
• Model Version selector now appears above the prompt for better workflow
• Advanced parameters reorganized by dependency — Vocal Gender, Persona ID, Persona Model only appear when Custom Mode is enabled
• Audio results now display Task ID and Result ID for easy copying

Gemini 3.1 Pro CustomTools is now available on EvoLink — a specialized Gemini 3.1 Pro variant optimized for advanced function calling and multi-step tool orchestration.

Gemini 3.1 Pro CustomTools

• Model ID: gemini-3.1-pro-preview-customtools
• Input: Text, image, video, audio, and PDF — up to 1,048,576 tokens
• Output: Text — up to 65,536 tokens
• Features: Enhanced function calling, multi-step tool orchestration, structured outputs with JSON schema validation, reasoning tokens tracking

Why CustomTools?

• Advanced tool use: Purpose-built for complex agentic workflows requiring reliable function calling
• Structured outputs: JSON schema validation ensures consistent, parseable responses
• Reasoning transparency: Track tool-planning overhead with reasoning token metrics
• Two API formats: OpenAI SDK (/v1/chat/completions) and Google Native API supported
• Gemini 3.1 Pro CustomTools API | API Docs

Sora 2 Pro has been upgraded to Sora 2 Pro Preview (sora-2-pro-preview) with a new per-second billing model and updated resolution options.

What's Changed

• Model ID: sora-2-pro → sora-2-pro-preview
• Billing: Changed from per-call to per-second pricing — pay only for the duration you need
• Duration options: 4s / 8s / 12s (previously 10s / 15s / 25s)
• Quality options: 720p / 1080p (previously Standard / HD), 1080p costs slightly more
• Try Sora 2 Pro Preview

Gemini 3.1 Flash Lite Preview is now available on EvoLink — the most cost-efficient model in the Gemini 3.1 family, built for high-volume, latency-sensitive workloads.

Gemini 3.1 Flash Lite Preview

• Model ID: gemini-3.1-flash-lite-preview
• Input: Text, image, video, audio, and PDF — up to 1,050,000 tokens
• Output: Text — up to 65,536 tokens
• Features: Function calling, structured outputs, thinking, code execution, context caching, Batch API, search grounding
• Pricing: Input $0.25/1M tokens, Output $1.50/1M tokens — significantly cheaper than Flash ($0.50/$3.00)

Why Flash Lite?

• Ultra-low cost: Ideal for translation, classification, data extraction, and simple agentic tasks at scale
• Full multimodal support: Same input modalities as Flash and Pro
• Two API formats: OpenAI SDK (/v1/chat/completions) and Google Native API supported
• Gemini 3.1 Flash Lite Preview API | API Docs

Kling Custom Element is now available on EvoLink. Create reusable subject elements from reference images or videos, then use them across Kling O3 and V3 Image-to-Video for consistent character appearance.

Kling Custom Element

• Model ID: kling-custom-element
• Input: Reference images (frontal + angles) or reference video (max 8s) — reference must contain a visible human face
• Output: Reusable element_id for video generation
• Compatible models: Kling O3 series (text/image/reference/edit), Kling V3 Image-to-Video, and Kling V3 Motion Control via element_list
• Pricing: 1 Credit per element creation, full refund on failure
• Kling O3 — Custom Element | Kling V3 — Custom Element | Kling V3 MC — Custom Element

Kling 3.0 Motion Control is now available on EvoLink. Transfer human motion from a reference video onto a character image with per-second billing from $0.1134/s.

Kling 3.0 Motion Control

• Model ID: kling-v3-motion-control
• Input: Reference image + reference video
• Resolution: 720p / 1080p
• Duration: 3–10s (image orientation) / 3–30s (video orientation)
• Features: Character orientation control, custom element support, original audio preservation
• Pricing: From $0.1134/s (per-second billing)
• Kling 3.0 Motion Control API

We are excited to announce the launch of MiniMax-M2.5, MiniMax's latest text model, now available on EvoLink.

MiniMax-M2.5

• Model ID: MiniMax-M2.5
• Context Window: 204K tokens
• Max Output: 131K tokens (reasoning chain + reply)
• Features: Built-in reasoning, web search, prompt caching
• Compatibility: 100% compatible with OpenAI SDK — just change the base URL

Pricing (up to 40% off official)

• Input: $0.18 / 1M tokens (official $0.30)
• Output: $0.72 / 1M tokens (official $1.20)
• Cache Hit: $0.024 / 1M tokens (official $0.03)
• Web Search: $0.00056 / search

Key Use Cases

• Coding Agents: Repo Q&A, code generation, bug triage, review workflows
• Research: Search-augmented research flows with turbo/max strategies
• Document Processing: Long-context summarization and extraction
• MiniMax-M2.5 API

We are excited to announce the launch of GPT-5.4, OpenAI's latest and most capable text generation model, now available on EvoLink.

GPT-5.4

• Model ID: gpt-5.4
• Features: Advanced text generation, reasoning, and code capabilities
• Highlights: OpenAI's most powerful model with improved performance across benchmarks
• Context Window: Extended context window for complex tasks
• Compatibility: 100% compatible with OpenAI SDK — just change the base URL

Key Capabilities

• Advanced Reasoning: Superior performance in math, logic, and complex problem-solving
• Code Generation: Enhanced coding capabilities across multiple programming languages
• Multilingual: Improved support for diverse languages
• Cost Savings: Access via EvoLink at competitive pricing
• GPT-5.4 API

We are excited to announce the launch of WAN 2.6 Flash video generation variants on EvoLink, offering faster processing for image-to-video and reference-guided workflows.

WAN 2.6 Image-to-Video Flash

• Model ID: wan-2.6-i2v-flash
• Features: Fast image-to-video generation with audio support
• Duration: 3-8 seconds
• Highlights: Optimized for speed, ideal for rapid iteration

WAN 2.6 Reference-Video Flash

• Model ID: wan-2.6-ref-flash
• Features: Reference-guided video generation with fast processing
• Duration: 3-8 seconds
• Highlights: Use reference images to guide video style and content
• WAN 2.6 API

Nano Banana 2 Beta has been integrated as a variant of Nano Banana 2.

Nano Banana 2 Beta

• Model ID: nano-banana-2-beta
• Now integrated as a variant of Nano Banana 2
• Built-in Web Search and Image Search grounding for real-world accuracy
• Supports 1K / 2K / 4K output quality options
• Nano Banana 2 Beta API

Nano Banana 2 has been updated to the gemini-3.1-flash-image-preview model with new quality tiers and tool support.

Nano Banana 2

• Model ID: gemini-3.1-flash-image-preview
• Added 1K / 2K / 4K quality output options
• Added Web Search tool support (web_search parameter)
• New size options: 1:4, 4:1, 1:8, 8:1
• Reference image limit increased to 14
• Nano Banana 2 API

Nano Banana Pro

• Model ID (Official): gemini-3-pro-image-preview
• Model ID (Beta): nano-banana-pro-beta
• Added Web Search tool support for Official variant (web_search parameter)
• Nano Banana Pro documentation has been fully separated into Official and Beta variants.
• Nano Banana Pro API
• Nano Banana Pro Beta API

We are excited to announce the launch of Seedream 5.0 Lite, ByteDance's latest image generation model, now available on EvoLink.

Seedream 5.0 Lite

• Model ID: doubao-seedream-5.0-lite
• Features: High-quality image generation with flexible aspect ratios and custom resolution
• Highlights: Supports both preset ratios (1:1, 2:3, 3:2, etc.) and custom pixel dimensions
• Resolution: 376px–1536px per side, total pixels 3,686,400–10,404,496
• Output Formats: JPEG, PNG, WebP

Key Capabilities

• Flexible Sizing: Choose from 9 preset aspect ratios or enter custom width × height
• High Quality: Advanced image generation with fine-grained detail
• Multiple Output Formats: JPEG (default), PNG, and WebP support
• Priority Control: Optional prompt priority for advanced users

Documentation

• Seedream 5.0 API Documentation

We are excited to announce the launch of Gemini 3.1 Pro Preview, the latest iteration in Google's Gemini 3 multimodal model family, now available on EvoLink.

New Model

Gemini 3.1 Pro Preview

• Model ID: gemini-3.1-pro-preview
• Modality: Text, Image, Video, Audio, PDF input → Text output
• Context Window: 1,048,576 tokens (1M)
• Max Output: 65,536 tokens

Key Capabilities

• Multimodal Reasoning: Accepts text, code, image, video, audio, and PDF inputs
• Thinking Mode: Deeper reasoning with thinking tokens
• Tool Use: Function calling, structured outputs, code execution, and file search
• Search Grounding: Google Search grounding + URL context
• Caching: Implicit and explicit caching supported
• Batch API: Long-context pipeline support

Pricing

Same pricing as Gemini 3 Pro — Input $2/1M tokens, Output $12/1M tokens, with >200K token tiered pricing. Up to 20% discount for registered users.

Documentation

• Gemini 3.1 Pro API Documentation
• Gemini 3.1 Pro Model Page

We are excited to announce the launch of Claude Sonnet 4.6 and Doubao Seed 2.0 series, now available through EvoLink API.

Claude Sonnet 4.6

Anthropic's latest Sonnet model from the Claude 4.6 family, delivering strong performance with optimized cost-efficiency.

• Model ID: claude-sonnet-4-6
• Context Window: 200K tokens
• Max Output: 128K tokens
• Features: Advanced reasoning, code generation, multilingual support, web search tool

Key Capabilities

• High-quality coding and technical analysis
• Complex reasoning with extended thinking support
• Multilingual understanding (7 languages supported)
• Prompt caching for cost optimization

Seed 2.0 Pro

ByteDance's flagship model with top-tier reasoning and generation capabilities.

• Model ID: seed-2-0
• Context Window: 256K tokens
• Max Output: 128K tokens
• Features: Advanced reasoning, code generation, content creation, multi-turn conversation

Seed 2.0 Lite

Balanced model optimized for cost-efficiency with strong performance.

• Model ID: seed-2-0-lite
• Context Window: 256K tokens
• Max Output: 128K tokens
• Features: General-purpose conversation, content generation, cost-effective

Seed 2.0 Mini

Lightweight model for fast response and high-throughput scenarios.

• Model ID: seed-2-0-mini
• Context Window: 256K tokens
• Max Output: 128K tokens
• Features: Fast inference, high throughput, lightweight tasks

Seed 2.0 Code

Code-specialized model with Pro-level performance for programming tasks.

• Model ID: seed-2-0-code
• Context Window: 256K tokens
• Max Output: 128K tokens
• Features: Code generation, debugging, code review, technical documentation

Documentation

• Claude Sonnet 4.6 API Documentation
• Seed 2.0 API Documentation

We are excited to announce the launch of Grok Imagine Video and DeepSeek series models, now available through EvoLink API.

Grok Imagine Video

xAI's video generation model supporting text-to-video and image-to-video with multiple creative modes.

• Model ID: grok-imagine-video
• Duration: 6–10 seconds
• Modes: Fun / Normal / Spicy
• Aspect Ratios: 16:9, 9:16, 1:1
• Input: Text prompt or image + text prompt

DeepSeek Chat

High-performance general-purpose chat model with 128K context window.

• Model ID: deepseek-chat
• Context Window: 128K tokens
• Features: General-purpose conversation, code generation, content creation

DeepSeek Reasoner

Advanced reasoning model with chain-of-thought capabilities, optimized for complex problem-solving.

• Model ID: deepseek-reasoner
• Context Window: 128K tokens
• Features: Chain-of-thought reasoning, math, logic, complex analysis

Documentation

• Grok Imagine Video API Documentation
• DeepSeek Chat API Documentation
• DeepSeek Reasoner API Documentation

We are excited to announce the launch of two new Kling video generation models: Kling O3 (V3 Omni) and Kling 3.0, offering powerful text-to-video, image-to-video, and more capabilities.

Kling O3 (V3 Omni)

Kling's next-generation unified video model with 4 generation modes, per-second billing from $0.075/s.

Text to Video

• Model ID: kling-o3-text-to-video
• Duration: 3–15 seconds
• Resolution: 720p / 1080p
• Aspect Ratios: 16:9, 9:16, 1:1

Image to Video

• Model ID: kling-o3-image-to-video
• Features: First frame, end frame, and reference images support
• Duration: 3–15 seconds
• Resolution: 720p / 1080p

Reference to Video

• Model ID: kling-o3-reference-to-video
• Features: Generate videos guided by reference video + reference images
• Duration: 3–10 seconds
• Resolution: 720p / 1080p
• Price: $0.1125/s (1.5x)

Video Edit

• Model ID: kling-o3-video-edit
• Features: AI-powered video editing and transformation
• Input Video: 3–10 seconds
• Resolution: 720p / 1080p
• Price: $0.1125/s (1.5x)

Kling 3.0

Kling 3.0 video model with text-to-video and image-to-video, per-second billing from $0.075/s.

Text to Video

• Model ID: kling-v3-text-to-video
• Duration: 3–15 seconds
• Resolution: 720p / 1080p
• Aspect Ratios: 16:9, 9:16, 1:1

Image to Video

• Model ID: kling-v3-image-to-video
• Features: First frame, end frame, subject control, multi-shot
• Duration: 3–15 seconds
• Resolution: 720p / 1080p

Documentation

• Kling O3 API Documentation
• Kling 3.0 API Documentation

We're excited to announce several major platform updates, including the new Claude Opus 4.6 model, Sora 2 Beta Max variant, and significant improvements to the image loading experience.

New Model: Claude Opus 4.6

The most capable model from Anthropic's Claude 4.6 series is now available on EvoLink.

• Model ID: claude-opus-4-6
• Features: Industry-leading reasoning, complex task handling, and extended context
• Highlights: Best-in-class performance for enterprise applications
• Context Window: 200K tokens
• Knowledge Cutoff: April 2024

Key Capabilities

• Advanced coding and technical analysis
• Complex mathematical reasoning
• Multilingual understanding (7 languages supported)
• Long-form content generation

Sora 2 Beta Max - Premium Video Generation

A new premium tier for Sora 2 with simplified pricing and enhanced features.

• Model ID: sora-2-beta-max
• Pricing: Duration-based only (no watermark fee)
• Resolution: 16:9 / 9:16
• Duration: 10s (1.0x) / 15s (1.5x)
• Features: Automatic watermark removal included

Pricing Advantage

• 10 seconds: Base pricing (1.0x multiplier)
• 15 seconds: 1.5x multiplier
• No additional fees: Watermark removal included at no extra cost

Sora 2 Preview - New Resize Mode Parameter

Added resize_mode parameter for better image-to-video generation control.

• Parameter: resize_mode
• Options:

  • pad: Add black bars to fit aspect ratio (default)
  • crop: Center crop to match aspect ratio
• Use Case: Handle aspect ratio mismatch when using input images

Platform Improvements

Enhanced Image Loading Experience

Improved handling of image loading failures with intelligent retry mechanism.

• Smart Detection: Automatically distinguish between network errors and expired URLs
• Retry on Network Errors: Users can retry failed images with one click
• Better Error Messages: Clear indication of temporary issues vs. permanent failures
• Multiple Status: loading → checking → network_error/expired → retry

User Experience Enhancements

• Fixed button group layout for 2-option selectors
• Improved image status indicators in Playground history
• Optimized error handling with detailed status tracking

Documentation

• Claude Opus 4.6 Documentation
• Sora 2 Beta Max Documentation
• Sora 2 Preview API Reference

We are excited to announce major updates to our video generation models, including 4K support for VEO 3.1 and 1080P support for Seedance 1.5 Pro.

VEO 3.1 Vertex AI - Now with 4K Support

VEO 3.1 Fast (Vertex AI)

• Model ID: veo-3.1-fast-generate-preview
• Features: Fast generation with 4K support, reference images
• Highlights: Balanced speed and quality, ideal for rapid iteration
• Resolution: 720p / 1080p / 4K
• Duration: 4s / 6s / 8s

VEO 3.1 Pro (Vertex AI)

• Model ID: veo-3.1-generate-preview
• Features: Highest quality with 4K support, reference images
• Highlights: Professional-grade output, best for final production
• Resolution: 720p / 1080p / 4K
• Duration: 4s / 6s / 8s

Seedance 1.5 Pro - Now with 1080P Support

• Model ID: seedance-1.5-pro
• New Feature: 1080P resolution now available
• Resolution: 480p / 720p / 1080p
• Duration: 4-12 seconds

Key Capabilities

• VEO 3.1 4K: Generate stunning 4K videos for professional use
• Seedance 1080P: Higher resolution output for Seedance 1.5 Pro
• Reference Images: Guide generation with reference images
• Flexible Duration: Multiple duration options for both models

Documentation

• VEO 3.1 API Documentation
• Seedance 1.5 Pro API Documentation

We are excited to announce the launch of Seedance 1.5 Pro, BytePlus's latest video generation model with integrated audio generation capabilities.

New Model

Seedance 1.5 Pro

• Model ID: seedance-1.5-pro
• Features: Text-to-Video and Image-to-Video generation with optional synchronized audio
• Highlights: Unified model supporting both T2V and I2V with AI-generated audio
• Resolution: 480p / 720p
• Duration: 4-12 seconds

Key Capabilities

• Text-to-Video: Generate videos directly from text descriptions
• Image-to-Video: Animate static images with motion
• Audio Generation: Optional synchronized audio including voice, sound effects, and background music
• Multiple Aspect Ratios: 16:9, 9:16, 1:1, 4:3, 3:4, 21:9, and Adaptive
• Reference Images: Support for 1-2 reference images for guided generation

Documentation

• Seedance 1.5 Pro API Documentation

We are excited to announce the launch of Suno AI music generation model, bringing you professional-quality AI-powered music creation capabilities.

New Model

Suno

• Model ID: suno-v5 (also supports suno-v4.5plus, suno-v4.5all, suno-v4.5, suno-v4)
• Features: AI-powered music generation with vocals, lyrics, and instrumental support
• Highlights: Create professional-quality songs from text prompts, multiple model versions
• Duration: Up to 4 minutes per song

Key Capabilities

• Simple Mode: Just provide a prompt, AI generates lyrics and melody automatically
• Custom Mode: Full control over style, title, and lyrics
• Instrumental Mode: Generate music without vocals
• Song Extension: Continue or extend existing songs

Documentation

• Suno API Documentation

We are excited to announce the launch of MiniMax Hailuo video generation model series, bringing you cutting-edge AI video creation capabilities.

New Models

Hailuo 2.3 Fast

• Model ID: MiniMax-Hailuo-2.3-Fast
• Features: Optimized for speed, Image-to-Video (I2V) mode only
• Highlights: Lowest latency, extreme physics effects, ideal for rapid iteration
• Resolution: 768P / 1080P
• Duration: 6s / 10s

Hailuo 2.3

• Model ID: MiniMax-Hailuo-2.3
• Features: High-quality output with Text-to-Video (T2V) and Image-to-Video (I2V) dual modes
• Highlights: SOTA instruction following, professional-grade quality
• Resolution: 768P / 1080P
• Duration: 6s / 10s

Hailuo 02

• Model ID: MiniMax-Hailuo-02
• Features: Full-featured with Text-to-Video (T2V), Image-to-Video (I2V), and First-Last-Frame (FLF) three modes
• Highlights: Supports 512P low-cost option, ideal for diverse creative needs
• Resolution: 512P / 768P / 1080P
• Duration: 6s / 10s

Key Capabilities

• 15 Camera Commands: Pan, Tilt, Zoom, Push, Pull, Truck, Pedestal, Static, Tracking, Shake and more
• Intelligent Mode Detection: System auto-detects mode based on input image count
• Prompt Optimizer: AI-powered prompt enhancement for better results
• Fast Pretreatment: Optional acceleration for image analysis

Documentation

• Hailuo 2.3 API Documentation
• Hailuo 02 API Documentation

Thank you for your support and feedback during the Phase 1 promotion. As the "Alibaba Model Full Series" Phase 1 promotion concludes, we are launching the Phase 2 pricing structure.

Promotion Details

• Phase 2 Pricing: Unified 30% off official prices (covering all Alibaba model series capabilities)
• Effective Date: December 26, 2025 00:00
• Scope: Alibaba model series (image/video generation, etc.) related billing items

Notes

After the new pricing takes effect, new API calls will be billed at Phase 2 promotional prices; historical bills remain unaffected.

1. GPT Image 1.5

   • gpt-image-1.5:

     • OpenAI's advanced image generation model.
     • Optimized for true-color precision and structured visual tasks.
     • Enhanced analytical visual outputs based on DALL-E 3 technology.
     • High-quality image generation with improved detail accuracy.
   • Documentation: GPT Image 1.5 API
2. Gemini 3 Flash Preview

   • gemini-3-flash-preview:

     • Google's latest fast inference model preview.
     • Optimized for speed while maintaining high quality output.
     • Extended context window support.
     • Ideal for high-throughput applications.
   • Documentation: Gemini 3 Flash Preview API

1. WAN 2.6 Text to Video

   • wan2.6-text-to-video:

     • Alibaba Tongyi Wanxiang 2.6 text-to-video generation.
     • Enhanced quality with 720p/1080p resolution support.
     • Multiple duration options: 5s, 10s, 15s.
     • Intelligent prompt rewriting for better results.
   • Documentation: WAN 2.6 Text to Video API
2. WAN 2.6 Image to Video

   • wan2.6-image-to-video:

     • Transform static images into dynamic videos.
     • First-frame generation with motion control.
     • Supports 5s, 10s, 15s duration at 720p/1080p quality.
   • Documentation: WAN 2.6 Image to Video API
3. WAN 2.6 Reference Video

   • wan2.6-reference-video:

     • Extract character appearance and voice from reference videos.
     • Generate new videos with consistent character identity.
     • Advanced video-to-video transformation capabilities.
   • Documentation: WAN 2.6 Reference Video API

1. OmniHuman 1.5 (digital human video)

   • omnihuman-1.5: AI-driven digital human video with audio-driven lip sync.
   • Turns a still portrait into a talking digital human with natural expressions and motion; supports 1:1, 9:16 and 16:9 aspect ratios.
   • Billed by audio duration; the pricing page is the source of truth for current rates.
   • Docs: OmniHuman 1.5 API
2. GPT-5.2 (text / multimodal)

   • gpt-5.2: OpenAI's flagship model tuned for coding and agent workloads.
   • 400K context window, up to 128K output tokens, and stronger reasoning for complex multi-step workflows.
   • Prompt caching is built in; input, output and cache hits are priced separately — the pricing page is the source of truth for current rates.
   • Docs: GPT-5.2 API

1. Z Image Turbo (image)

   • z-image-turbo: an ultra-fast image generation model built on Tongyi Lab's open architecture.
   • Around 3 seconds per generation, up to 4K resolution, suited to high-concurrency batch workloads, with strong prompt fidelity.
   • The pricing page is the source of truth for current rates.
2. Key capabilities

   • Multiple aspect ratios: 1:1, 9:16, 16:9 and more
   • Negative prompt control
   • NSFW content filter toggle
   • Custom LoRA model loading

Docs: Z Image Turbo

1. Kling O1 series (video)

   • kling-o1-image-to-video: image-to-video with up to 2 input images, 5 or 10 seconds.
   • kling-o1-video-edit: AI video editing with reference images for style guidance, 3-10 seconds.
   • kling-o1-video-edit-fast: fast video editing mode, 6-20 seconds.
   • Docs: Kling O1 image-to-video | Kling O1 video edit | Kling O1 fast video edit
2. Seedream 4.5 (image)

   • doubao-seedream-4.5: stronger multi-image editing consistency, improved text rendering in images, up to 4K resolution.
   • Docs: Seedream 4.5

The three Kling O1 models bill per second and Seedream 4.5 bills per image; the pricing page is the source of truth for current rates.

1. Nano Banana 2 series

   • nano-banana-2-lite: size and resolution configuration are now supported (1K / 2K / 4K).
   • nano-banana-2-pro (direct official channel): on top of size configuration, a 4K resolution option is now available.
   • Image input and image output are metered separately, and resolutions form separate price tiers; the pricing page is the source of truth for current rates.
   • Docs: Nano Banana 2
2. New Veo 3.1 Fast capabilities

   • Multi-image reference: up to 3 reference images, at no change in price.
   • Video extend: extend a video from a prompt plus a previously generated video ID, via model veo3.1-fast-extend, priced the same as veo3.1-fast.
   • Docs: Veo 3.1 Fast

Update July 3, 2026: Nano Banana 2 Lite now refers to Google's Gemini 3.1 Flash-Lite Image route on EvoLink. Earlier launch wording should be treated as historical context, not the current model status.

1. Two versions

   • gemini-3-pro-image-preview: the official channel, more stable and controllable, for demanding commercial work.
   • nano-banana-2-lite: the Google Gemini 3.1 Flash-Lite Image route, aimed at cost-effective 1K image generation.
2. Model naming

   • To protect quality, gemini-3-pro-image-preview points at the official version by default.
   • To switch to Lite, change the model name to nano-banana-2-lite — usually a one-minute change.
3. What comes next

   • Pro will progressively gain multi-image and 4K capabilities.
   • Lite is onboarding more channels to improve capability and stability.

The two versions are priced differently; the pricing page is the source of truth for current rates.

• Nano Banana 2 is live on EvoLink, backed by the official model gemini-3-pro-image-preview.
• Docs: Nano Banana Pro
• The pricing page is the source of truth for current rates.

1. Sora 2 Pro available

   • Priced in tiers by video duration and resolution; the pricing page is the source of truth for each tier.
2. Gemini 3 status

   • Integration is complete, but test results do not yet meet our bar, so external calls stay closed.
   • We will notify users as soon as it qualifies.

1. Supported models

   • claude-haiku-4-5-20251001
   • claude-sonnet-4-5-20250929
   • claude-opus-4-1-20250805
   • claude-sonnet-4-20250514
2. Use cases and docs

   • Engineered for coding workloads, with large customers already running on it.
   • The pricing page is the source of truth for current rates. Docs: Claude Messages API

1. New models

   • Gemini 2.5 Flash is available.
   • Seedance 1.0 Pro Fast is available, with 720P and 1080P resolution tiers.
   • The pricing page is the source of truth for current rates.
2. Platform improvements

   • Payments: a new Credits page improves the top-up and checkout flow.
   • Sora 2: the prompt character limit is raised to 5,000.
   • Logs: call logs now explain status, generated result and failure details.

1. Sora 2 duration

   • Sora 2 now generates 15-second videos.
   • Watermarked and watermark-free are two separate price tiers; the pricing page is the source of truth for current rates.
2. Platform updates

   • Webhook: callback parameters are now supported.
   • Website: pages are available in multiple languages, no browser translation plugin needed.
3. Veo 3 sunset notice

   • Nearly all traffic has moved to veo3.1-fast, so the legacy Veo 3 will be retired in 3 days.
   • If you still call it, change the model parameter to "veo3.1-fast".

1. New channel

   • A new channel that passed our stability tests is now serving Veo 3.1.
2. Smart routing

   • The system picks an available channel automatically; callers do not need to change anything.
   • If the preferred channel is unavailable, traffic switches to a backup channel to preserve stability.
3. Next steps

   • We keep evaluating and onboarding channels that are both stable and cost-effective.

The pricing page is the source of truth for current rates.

1. Capabilities

   • Text-to-video, image-to-video, and video generation from custom first/last frames.
   • Docs: Veo 3.1 Fast
2. Channels and roadmap

   • The channel in use is stable and matches official capability.
   • Extended-length video (video extension) is in integration testing and will open once it meets our bar.

The pricing page is the source of truth for current rates.

1. Why this model

   • Direct official channel: stable, low latency, no quality degradation, multimodal input supported.
   • Docs: Gemini 2.5 Pro quickstart
2. Concurrent promotion

   • wan 2.5 video, wan 2.5 image and Qwen Image Edit are on a limited-time promotional rate.
   • The promotion is time-limited and we will give notice before it ends; the pricing page is the source of truth for current rates.

1. Parameter support

   • remove_watermark controls the watermark: true removes it, false keeps it.
   • The parameter defaults to true: omit it and the request is treated as the watermark-free version, which affects billing.
2. Billing

   • Watermark-free and watermarked are two separate price tiers; the pricing page is the source of truth for current rates.
   • See the latest Sora 2 docs for details.

• Platform pricing has been updated. The pricing page and the individual model pages are the source of truth for current rates.
• Sora 2 will move to tiered pricing based on generation parameters (duration, resolution); parameters and rates are updated together on the model page.
• If upstream providers or channels force a change, we will announce it in advance.

• Added horizontal screen and vertical screen parameter selection to support different screen ratios.
• Supports uploading reference images, please see the document for specific parameters and calling methods: Sora 2 Document

Risk control tips (reference picture)

• OpenAI official review of reference images is extremely strict.
• Reference pictures containing real-life photos are likely to fail in the review. It is recommended to give priority to non-real-life pictures such as animations, landscapes, and concept pictures to increase the success rate.

• Sora channels are fully restored; the earlier instability has been traced and fixed.
• From October 5, Sora moves to regular billing. The pricing page is the source of truth for current rates.
• We will keep improving the interface based on what we learned over the past few days, for better stability and experience.

• The Sora 2 API is now open. If you already call another video model, switching to Sora 2 takes minutes — see the Sora 2 docs.
• Note: Sora 2 takes about 5 minutes to generate; leave enough headroom in your business logic.
• The pricing page is the source of truth for current rates.

Sora 2 stability and content filtering

• Overnight monitoring put the Sora success rate around 88.24%; channels overall are still limited by the lack of an officially stable upstream interface.
• Current failures concentrate in content moderation. Sora filters aggressively — prompts with mild sexual or horror elements fail easily.
• Refine your prompts; if the failure rate stays high after ruling out moderation, contact us and we will help investigate.

New models:

1. Image: qwen-image-edit, wan2.5-text-to-image, wan2.5-image-to-image. Docs: Wan 2.5 text-to-image
2. Video: wan2.5-text-to-video, wan2.5-image-to-video. Docs: Wan 2.5 text-to-video

Seedream size parameter now accepts a wider range of values — see the Seedream docs.

Coming soon: Gemini 2.5 Pro and Claude 4.5, on official channels.

The pricing page is the source of truth for per-model rates.

System update today.

GPT-4O-Image model update:

1. Supported sizes: 1024x1024 (1:1), 1024x1536 (2:3), 1536x1024 (3:2).
2. Batch generation: 1, 2, 4 pictures.
3. Documentation: GPT-4O Model Page

SeeDream-4.0 model update:

1. Size: simplified (1K, 2K, 4K); precise (9 ratios including 1024x1024).
2. Batch generation: 1-15 pictures.
3. Documentation: SeeDream Documentation

Thanks for the user suggestions and looking forward to feedback.

We observed stability fluctuations on the main Nano Banana channel. Here is where things stand.

Channel notice:

1. The system routes automatically to a currently available channel; callers do not need to change their requests.
2. After stability dropped on some channels, traffic was shifted to better-performing ones.
3. We are in contact with channel providers and testing new channels; we will share progress as it comes.

The platform pricing policy has been adjusted. The pricing page and the individual model pages are the source of truth for current rates; any change driven by upstream providers will be announced in advance.

This update strengthens API key management.

API key security improvements:

1. Set a quota cap and the range of callable models when creating a key.
2. Usage statistics show consumed and remaining quota.
3. Adjust the quota of an existing key without rotating it.

Thanks for the feedback — suggestions are always welcome.

The EvoLink.AI platform is open for use, providing developers with stable model API services.

Models available at launch:

• Seedream 4.0 — image generation
• Nano Banana — image generation
• Veo3-Fast — video generation
• GPT-4o Image — image generation

The GPT-4o Image model page and pricing page are the source of truth for current rates and parameters.