Gemini 3.0 Pro Preview API

Access gemini-3-pro-preview through EvoLink's OpenAI-compatible /v1/chat/completions endpoint. Built for coding, multimodal reasoning, and tool use with detailed token accounting.

Using coding CLIs? Run Gemini 3 Pro via EvoCode — One API for Code Agents & CLIs. (View Docs)

PRICING

PLAN	CONTEXT WINDOW	MAX OUTPUT	INPUT	OUTPUT	CACHE READ
Gemini 3 Pro	1.05M	65.5K	≤200.0K$1.60-20% $2.00Official Price >200.0K$3.20-20% $4.00Official Price	≤200.0K$9.60-20% $12.00Official Price >200.0K$14.40-20% $18.00Official Price	≤200.0K$0.160-20% $0.200Official Price >200.0K$0.319-20% $0.400Official Price
Gemini 3 Pro (Beta)	1.05M	65.5K	≤200.0K$0.520-74% $2.00Official Price >200.0K$1.04-74% $4.00Official Price	≤200.0K$3.12-74% $12.00Official Price >200.0K$4.68-74% $18.00Official Price	≤200.0K$0.052-74% $0.200Official Price >200.0K$0.104-74% $0.400Official Price

Pricing Note: Price unit: USD / 1M tokens

Cache Hit: Price applies to cached prompt tokens.

Two ways to run Gemini 3 Pro — pick the tier that matches your workload.

· Gemini 3 Pro: the default tier for production reliability and predictable availability.
· Gemini 3 Pro (Beta): a lower-cost tier with best-effort availability; retries recommended for retry-tolerant workloads.

Gemini 3.0 Pro Preview API — OpenAI SDK Compatible

Call gemini-3-pro-preview using the OpenAI SDK format via EvoLink. Get usage plus reasoning_tokens in responses and leverage a 1M context window reported in testing.

Capabilities of Gemini 3.0 Pro Preview API

Multimodal Inputs + Grounding

Gemini 3 Pro Preview API accepts text, code, image, video, audio, and PDF inputs with text-only output, and supports Search grounding plus URL context for verifiable answers.

Explore Multimodal

Thinking + Agent Tools

Thinking, function calling, structured outputs, code execution, and file search are supported for agentic reasoning and automation.

Build Agents

1M Context + Ops Controls

Input token limit is 1,048,576 with up to 65,536 output tokens. Caching and Batch API support long-context pipelines.

Check Limits

Why Use Gemini 3.0 Pro on EvoLink

EvoLink exposes Gemini 3.0 Pro Preview through a familiar OpenAI SDK-style endpoint, with explicit auth and detailed usage stats for production-grade tracking.

OpenAI SDK Format

Call /v1/chat/completions using model + messages. The messages array is required (minimum length 1).

Granular Usage Metrics

Usage includes prompt_tokens, completion_tokens, total_tokens, plus completion_tokens_details.reasoning_tokens.

Model Quality Signal

Vercel reports stronger instruction following, improved response consistency, and strong results in its Next.js evaluations.

How to Call Gemini 3.0 Pro Preview

Use the OpenAI SDK format and the gemini-3-pro-preview model string.

Step 1 - Set the Model

Use model: "gemini-3-pro-preview" in the request body.

Step 2 - Send Messages

Provide a messages array with role/content pairs (minimum length 1).

Step 3 - Inspect Output + Usage

Read choices[0].message.content and track usage.prompt_tokens, completion_tokens, and reasoning_tokens.

Open Developer Playground

Technical Specs

Key details for the Gemini 3.0 Pro Preview API

Compatibility

OpenAI SDK Format

Use the standard /v1/chat/completions interface.

Model

Model String

Set model to gemini-3-pro-preview for this endpoint.

Analytics

Usage Breakdown

Response includes prompt/completion totals plus detailed token categories.

Reasoning

Reasoning Tokens

completion_tokens_details includes reasoning_tokens for deeper analysis.

Multimodal

Multimodal Reasoning Focus

Vercel notes stronger multimodal reasoning and tool use in testing.

Context

1M Context Window

Vercel reports a 1M context window supporting long agent flows.

Gemini 3.0 Pro API FAQs

Everything you need to know about the product and billing.

Gemini 3 is Google's latest multimodal model family. This page focuses on Gemini 3 Pro Preview (model code: gemini-3-pro-preview), which supports text, image, video, audio, and PDF inputs with text output, plus thinking, function calling, structured outputs, code execution, file search, search grounding, URL context, caching, and Batch API support. Google lists a January 2025 knowledge cutoff for this preview model.

Use "gemini-3-pro-preview" as the model value for the Gemini 3 Pro Preview API on EvoLink. This is the recommended model string for the OpenAI-compatible chat completions endpoint.

Send a POST request to https://api.evolink.ai/v1/chat/completions. Include Authorization: Bearer <token> and Content-Type: application/json using an API key from your EvoLink dashboard.

OpenAI SDK chat completions format with model and a required messages array (minimum length 1). Each message uses role/content pairs, multimodal inputs can be passed as content parts, and common options include stream, max_tokens, and temperature.

Inputs include text, image, video, audio, and PDF, and the output is text only. Image or audio generation is not supported in this Gemini 3 Pro Preview model.

Input token limit is 1,048,576 (1M context) and max output tokens are 65,536. This supports long documents and multi-turn contexts.

Thinking, function calling, structured outputs, code execution, file search, search grounding, URL context, caching, and Batch API are supported. For URL context, function calling with tool use is currently unsupported, and you can pass up to 20 URLs with 34MB max per URL.

EvoLink responses include usage.prompt_tokens, completion_tokens, total_tokens, plus prompt_tokens_details and completion_tokens_details.reasoning_tokens. Use these fields to monitor cost, latency, and reasoning token consumption.