Access MiniMax M3.1 Flash through EvoLink’s unified API gateway. Evaluate long-context coding, tool use and visual frontend workflows before choosing the model for your application.
Production routeThis rate reflects platform-side availability — only confirmed server errors (HTTP 500 / empty response) count as failures. User-side issues (content moderation, invalid params, cancellation) plus rate limits, timeouts and auth errors are excluded. Before real traffic arrives, empty buckets may display as available.Live
Live
Context
1M context · 524K max output
Best For
Agentic coding, tool-calling loops and long-context tasks
Agentic coding, tool use and long-context tasks on one model ID, MiniMax-M3.1-Flash. Input, cached input and output are billed separately; current rates are shown below.
MiniMax M3.1 Flash
MiniMax’s M-series Flash model for agentic coding, with tunable thinking depth
Selected
Model ID
MiniMax-M3.1-Flash
Best for
Coding agents, tool-calling loops and long-context tasks with a 1M context window; set an effort level per task and measure cost per accepted result on your own workload.
Input
$0.150 / 1M
10.2 cr / 1M
Cache hit
$0.015 / 1M
1 cr / 1M
Output
$0.600 / 1M
40.8 cr / 1M
All rates are per 1M tokens, shown in USD and credits, and reflect your account's current pricing.
MiniMax M3.1 Flash pricing
Estimate what one MiniMax M3.1 Flash request costs before you integrate. The calculator uses your account's current rates, with official pricing as a reference.
MiniMax M3.1 Flash is a multimodal reasoning model for coding agents, tool use and long-context work. Its 1M-token context and adjustable thinking depth make it a candidate for multi-file debugging, visual frontend work and research workflows. Evaluate it alongside other models through EvoLink’s unified API gateway, with the model configuration and current rates shown above.
MiniMax M3.1 Flash is served on EvoLink under the model ID MiniMax-M3.1-Flash through Chat Completions · Anthropic Messages · Responses, with the same API key and balance you use for every other model. It offers a 1M context window and up to 524K output tokens, plus multi-step agents, tool use, image input.
Numbers come from the EvoLink route configuration; capabilities are what the API exposes today.
Context window
1M tokens
Max output
524K tokens
Input
Text + images + video
Output
Text
Reasoning
Configurable reasoning effort
Tool use
Function calling with multi-step tool sequences
Prompt caching
Automatic cache reads at a lower rate
Protocols
Chat Completions · Anthropic Messages · Responses
Model ID
MiniMax-M3.1-Flash
Image input
JPEG · PNG · GIF · WebP
Video input
MP4 · AVI · MOV · MKV
The API model ID is MiniMax-M3.1-Flash, written with capitals, a dot and hyphens; this page’s URL uses lowercase hyphens. MiniMax-M3 is a separate model with its own rates.
MiniMax M3.1 Flash API and Preview: what should you check?
Search results cover different access channels and release stages. Use the current model configuration for your chosen service.
The API name and the Preview guide
MiniMax’s current API documentation lists MiniMax-M3.1-Flash. Its MiniMax Code guide uses MiniMax-M3.1-Flash-Preview. The guide’s examples describe the preview environment; a name change alone does not establish general availability or identical behavior across services.
A model API and a coding application
Use an API to add the model to your own application or agent. MiniMax Code supplies its own workspace and tools. Through EvoLink, you can evaluate model choices within one API gateway; file access, code execution and other agent tools come from the application you connect.
What can you build with the MiniMax M3.1 Flash API?
Use these tasks to plan an evaluation. MiniMax’s published preview guide provides workflow examples; outcomes depend on the supplied materials, tools and validation.
Find and fix bugs across a repository
Give a coding agent the relevant source files, a reproduction and acceptance criteria. Evaluate whether it traces the cause, makes a focused change and checks regressions, including edge cases that a plausible patch can miss.
Investigate frontend performance
MiniMax’s published Flowdesk case covers slow lists, search and pagination. In a comparable workflow, connect browser profiling tools and compare measurements before and after the change while checking that existing interactions still work.
Turn visual references into interactive frontend code
Combine screenshots or reference images with layout and interaction requirements. The preview guide includes a procedural Three.js reconstruction; test the resulting code for responsive layout, controls and behavior in your own browser environment.
Analyze image differences with tools
Ask the model to identify meaningful differences between two images. A connected image-processing tool can produce an annotated comparison, as shown in the preview guide. That workflow combines visual understanding with externally executed code.
Review videos and screen recordings
Use video input to examine scene changes, visible actions or a recorded interface problem, then request a structured text summary. Precise frame extraction, transcription and media editing require the relevant tools in your application.
Research long documents and financial workbooks
Work from source documents, explicit assumptions and a defined deliverable. MiniMax’s preview examples include financial modeling and supply-chain research; web research and spreadsheet creation need connected tools, and calculations and sources need independent review.
How do you choose between MiniMax M3.1 Flash and MiniMax-M3?
Both models stay available on EvoLink with separate model IDs and rates. Replay the same requests through each before moving traffic.
Thinking is always on in M3.1 Flash
M3.1 Flash cannot turn thinking off: requests that disable it are rejected with a 400 error. Control depth with the effort level instead — low, medium, high, xhigh or max. MiniMax-M3 can skip thinking entirely.
Effort defaults to max
If you send no effort level, M3.1 Flash thinks at max depth, which produces the most output tokens and the longest latency. Use low or medium for routine calls and keep the higher levels for hard tasks.
The effort field differs per protocol
Chat Completions uses reasoning_effort, Messages uses output_config.effort and Responses uses reasoning.effort. On Chat Completions the thinking text comes back in reasoning_content, separate from content.
One rate across the whole context window
On EvoLink, M3.1 Flash bills input, cached input and output at one rate regardless of prompt length, while MiniMax-M3 has a separate long-context tier. Compare both Pricing sections for current rates.
How do you keep MiniMax M3.1 Flash API cost predictable?
The Pricing section above holds the live rates and the request calculator. Use it as an estimate, then reconcile reported usage with final charges and accepted results.
Pick the lowest effort that passes
Thinking tokens are billed as output. Start each task type at low or medium and raise the effort level only where accepted results actually improve.
Leave room in the output limit
Thinking counts toward max_tokens. A limit that is too small can end with finish_reason length and an empty answer, and the thinking is still billed. Set limits per task and detect truncation.
Keep the prefix stable so it caches
Caching is automatic; explicit cache_control markers are not supported. Put reusable instructions and tool definitions first and read the cached tokens reported in usage instead of assuming hits.
Measure images and video on your own content
Image tokens vary with size and the detail setting, and video tokens with length and frame rate. Run a representative sample and use the reported input tokens in the calculator.
Two ways to use MiniMax M3.1 Flash: EvoLink API or Agent
Use the EvoLink API for product backends and batch jobs, or call MiniMax M3.1 Flash from Codex, Claude, or Gemini for coding and analysis workflows. Both paths share the same EvoLink API key, balance, model ID, and request history.
Option 1
Integrate with the EvoLink API
Best for: product backends, batch jobs, automated pipelines
Send OpenAI-compatible Chat Completions (or Anthropic Messages) requests to EvoLink and control the model ID, system prompt, output budget, tools, and structured output.
1Create an EvoLink API key in the console
2Point your OpenAI or Anthropic SDK at the EvoLink base URL and select the model ID shown above
3Send one representative request and read the usage field for input, cached, and output tokens
4Set max_tokens and retries per task; keep tool-call IDs and results across turns
Best for: coding, review, and analysis tasks in Codex, Claude, and Gemini
Give the Agent the task, the inputs to include, and the acceptance criteria. It assembles the request, calls MiniMax M3.1 Flash through EvoLink, and returns the answer with token usage.
1Set EVOLINK_API_KEY in your local environment; never put it in code or a prompt
2Describe the task, the inputs to include, and the expected output format
3Ask the Agent to call MiniMax M3.1 Flash through EvoLink and show the request before sending
4Let the Agent report the answer, token usage, and any error body
MiniMax M3.1 Flash API code example and error handling
This example shows the shortest runnable request: an OpenAI-compatible Chat Completions call with a system prompt, a user message, and an output budget. Open the API tab for the complete parameter and response reference.
curl -X POST https://api.evolink.ai/v1/chat/completions \
-H "Authorization: Bearer $EVOLINK_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "MiniMax-M3.1-Flash",
"messages": [
{ "role": "system", "content": "You are a senior engineer reviewing a pull request." },
{ "role": "user", "content": "Review this diff and list every risky change with file and line:\n<diff>" }
],
"reasoning_effort": "medium",
"max_tokens": 16384
}'
# Thinking is always on and counts toward max_tokens; omitting
# reasoning_effort means "max". Anthropic Messages (/v1/messages) and
# Responses (/v1/responses) accept the same model ID.
# The response returns thinking in choices[0].message.reasoning_content,
# the answer in content, and a usage object.
Invalid request or unsupported parameter
Check the model ID, the messages array, and parameter ranges against the API reference; remove fields this route does not support.
Authentication or balance issue
Check the Authorization bearer token and confirm the available balance in the console.
Context length exceeded
Prompt tokens exceed the MiniMax M3.1 Flash context window. Trim or retrieve only the relevant evidence and reuse cached prefixes.
Rate limited (429)
Back off and retry with jitter; batch or queue requests instead of sending parallel bursts.
Content or tool call rejected
Review sensitive content, malformed tool-call arguments, and JSON schema mismatches before retrying.
How do you send your first MiniMax M3.1 Flash request?
Four steps from an API key to production traffic.
01
Copy the exact model ID
Use MiniMax-M3.1-Flash, with capitals, a dot and hyphens exactly as written. The copy button on the Models card gives the exact string; the page URL uses a different form.
Model configuration
02
Send a text request with an effort level
Start on the protocol your stack already speaks, set an explicit effort level and output limit, and confirm the answer, the thinking field and usage on a small request.
First request
03
Add tools, images or video
Return thinking content unchanged in multi-turn tool conversations. For media, use image_url or video_url parts on Chat Completions, or image and video blocks on Messages, then compare usage with the text-only task.
Tools and media
04
Stabilise the prefix, then scale up
Keep reusable context stable, measure cache hits instead of assuming them, and increase traffic once quality, latency and final charges pass your acceptance criteria.
Gradual rollout
Cache hits are billed at a fraction of fresh input
MiniMax caches repeated prompt prefixes automatically and reports the hits in usage; explicit cache_control markers are not supported. Keep instructions and tool definitions first and identical across turns. A cache hit is not guaranteed for every request.
1M-token context with a 524K output ceiling
Thinking and the final answer share the output ceiling, so budget for both. Treat the window as capacity rather than a target: send the evidence the task needs and keep the prefix stable so it caches.
Compare MiniMax M3.1 Flash with MiniMax-M3 and DeepSeek V4.1 Flash
EvoLink
Compare input/output rates, context, caching and workload fit. Prices below follow your account’s current rates; benchmark the same requests before choosing a route.
Coding agents, tool-calling loops and long-context tasks with a 1M context window; set an effort level per task and measure cost per accepted result on your own workload.
Agentic and multimodal work where simple calls can skip thinking
Long-context coding and agent loops that also need to read images
MiniMax M3.1 Flash
Input / output$0.15 / $0.6
Context1M
CachingAutomatic cache hits
Best forCoding agents, tool-calling loops and long-context tasks with a 1M context window; set an effort level per task and measure cost per accepted result on your own workload.
Best forLong-context coding and agent loops that also need to read images
What should you confirm before moving MiniMax M3.1 Flash to production?
Verify output quality, supported controls and actual charges on representative traffic.
The model ID, protocol and effort are wired correctly
Confirm requests send MiniMax-M3.1-Flash with an explicit effort level and output limit, and that usage reports input, cached and output tokens separately.
No request still tries to disable thinking
Search your code and SDK defaults for thinking type disabled or an effort of none; on M3.1 Flash those return 400. Use the low effort level instead.
Quality holds on your existing prompts
Replay a fixed set of real tasks through MiniMax-M3 and M3.1 Flash and compare instruction following, tool results and accepted output, rather than assuming prompts carry over.
Latency and errors behave under load
Higher effort levels mean longer responses. Watch time to first token and total latency on representative traffic, and prepare for rate limits and timeouts.
Start with representative tasks, a moderate effort level and a small traffic share. Expand when quality, latency and cost per accepted result meet your bar, and keep a verified fallback model in configuration rather than in code.
MiniMax Model Family
Same API key and balance — switch tiers without changing your integration.
MiniMax M3.1 Flash
Current model
MiniMax’s M-series Flash model for agentic coding, with tunable thinking depth
MiniMax-M3
MiniMax’s 1M-context multimodal model, with thinking you can switch off
MiniMax’s current API documentation uses MiniMax-M3.1-Flash, while its MiniMax Code guide uses MiniMax-M3.1-Flash-Preview. These sources describe different access contexts. Check the model configuration for the service you use; preview examples alone do not establish identical behavior or availability on every route.
Do I need MiniMax Code to use a model API?
MiniMax Code is an application with its own workspace and tools. A model API lets you build the model into your own software or agent. EvoLink provides a unified gateway for model access; your application supplies the file access, execution environment and tools needed for the task.
Can MiniMax M3.1 Flash generate images or videos?
Its documented model output is text. It can understand visual input and help write code or direct connected tools. The images, videos, presentations and workbooks shown in an agent workflow may be produced by those tools; they are not evidence of native media output from the language model.
Is MiniMax M3.1 Flash open source or available to download?
The sources linked on this page document model access and preview workflows; they do not establish a downloadable open-weight release for this exact version. Check for an official checkpoint and license before planning self-hosting.
Does Flash mean every task will finish faster?
Flash is the model’s name. Completion time also depends on the task, context, thinking depth, tool execution and service load. Compare the same real tasks with your current model and measure useful completed results; a preview speed figure does not predict end-to-end agent latency.
What is the MiniMax M3.1 Flash model ID?
Use MiniMax-M3.1-Flash on EvoLink, written exactly as shown with capitals, a dot and hyphens. This page lives at /minimax-m3-1-flash; the URL form is not a model ID.
Can I turn thinking off?
No. M3.1 Flash always thinks; sending thinking type disabled or an effort of none returns a 400 error. To reduce latency and thinking tokens, lower the effort level instead.
Which effort levels are available, and what is the default?
low, medium, high, xhigh and max. If you omit the effort level, M3.1 Flash uses max. Chat Completions takes reasoning_effort, Messages takes output_config.effort and Responses takes reasoning.effort.
How are thinking tokens billed?
Thinking tokens count as output tokens: they are billed at the output rate and count toward your max_tokens limit. Check usage to see how much each effort level adds on your tasks.
Which protocols can I call it on?
The same model ID works on Chat Completions (/v1/chat/completions), Anthropic Messages (/v1/messages) and Responses (/v1/responses). Validate streaming, tool calls and usage on the protocol you choose before production.
What input does it accept?
Text, images (JPEG, PNG, GIF, WebP) and video (MP4, AVI, MOV, MKV). Image and video tokens are billed at the input rate; there is no separate media charge.
How does caching work?
Caching is automatic: repeated prompt prefixes can be served from cache and are billed at the lower cached-input rate when usage reports them. Explicit cache_control markers are not supported, and a cache hit is not guaranteed.
How is it different from MiniMax-M3?
Both have a 1M-token context window and text, image and video input. M3.1 Flash always thinks and adds five effort levels; MiniMax-M3 can skip thinking. On EvoLink they have separate model IDs and rates, so compare both Pricing sections.
