Gemini 3.1 Pro
⧉Gemini 3.1 Pro is the next-generation flagship multimodal model from Google, with a very long context window and top-tier reasoning and coding. Fully OpenAI-API compatible: set model to gemini-3.1-pro in your existing SDK. Low-reasoning tier by default (faster and cheaper); pass reasoning_effort=high for the high tier (billed at the Preview rate). Responses carry a reasoning_content field so you can see the thinking (streamed chunk by chunk as well).
What is the Gemini 3.1 Pro API and what does it cost?
Gemini 3.1 Pro is a Google Gemini model (gemini-3.1-pro-preview in Google's API, still in preview) that accepts up to 1,048,576 input tokens and returns up to 65,536 output tokens. KovAPI serves it as gemini-3.1-pro through an OpenAI-compatible API. On KovAPI it costs $0.50 per 1M input tokens and $3.00 per 1M output tokens (Google list price: $2 / $12 for prompts up to 200K tokens). Base URL: https://kovapi.com/v1 (OpenAI-compatible).
Measured performance
Real streaming requests sent through the public KovAPI API. Test window 2026-10-27 18:17 – 2026-10-27 18:19 UTC; sample size 100.
Output speed is not shown: here the answer arrives in one burst after the model finishes thinking, and the reported output tokens include the thinking, so there is no answer-streaming window to time. The full-response time above is the useful number.
Failed requests in this sample: HTTP 503 × 99; HTTP 502 × 1. Failed calls are not billed.
How this was measured
POST /v1/chat/completions with stream: true, one at a time with a 1-second gap, rotating through 10 prompts that each ask for about 200 words on a software topic, max_tokens 2500, reasoning_effort not sent (model default).
Time to first token runs from sending the request to the first streamed token of any kind (answer or reasoning text).
Output speed is answer tokens divided by the time from the first to the last answer token; reasoning tokens are subtracted when the API reports them, and a request is left out when the token count cannot be separated from hidden thinking.
Success means HTTP 200 and a completed stream with a non-empty answer. Percentiles are nearest-rank over successful requests. The test key belongs to an internal account, so these calls are not part of the public success-rate figures elsewhere on the site.
Measurements are repeated every week and this page updates itself; the date above is the latest run.
Cost example
KovAPI figures use the live price on this page; vendor figures use the vendor's published list price (sources below).
1,000 analysis calls
Each call sends 20,000 input tokens and receives 2,000 output tokens (thinking included), with reasoning_effort omitted.
| Where | Cost | Difference |
|---|---|---|
| KovAPI | $16.00 | — |
| Google Gemini API (prompts up to 200K tokens) | $64.00 | KovAPI 75% cheaper |
Price tiers on KovAPI: reasoning_effort omitted or low — $0.50 / $3.00 per 1M input / output tokens; reasoning_effort medium, high or xhigh — $0.50 / $3.00 per 1M input / output tokens.
Thinking tokens are billed as output on both sides (Google's price table reads 'Output price (including thinking tokens)'). Google charges $4 / $18 per 1M for prompts over 200K tokens; KovAPI has no long-context tier.
Vendor list prices checked 2026-09-25: Gemini API pricing, Gemini API: thinking.
Parameters & capabilities
"Vendor spec" is what the model's maker publishes. "Via KovAPI" is what our own requests through the gateway showed (last checked 2026-10-27).
| Item | Vendor spec | Via KovAPI |
|---|---|---|
| Model id | gemini-3.1-pro-preview Google: Gemini 3.1 Pro model page |
gemini-3.1-pro |
| Endpoints | — | /v1/chat/completions |
| Price per 1M tokens | $2 input · $12 output (prompts up to 200K); $4 · $18 above 200K Gemini API pricing | $0.50 input · $3.00 output at any prompt length (reasoning_effort omitted or low) |
| Input limit | 1,048,576 tokens Google: Gemini 3.1 Pro model page | Not measured by us |
| Max output | 65,536 tokens Google: Gemini 3.1 Pro model page | ✗ max_tokens was not enforced in our test |
| Streaming | — | Not tested |
| Tool calling | — | ✗ Did not return a tool call in our test |
| Image input (vision) | — | ✗ Failed in our test |
| JSON mode | — | ✗ Not enforced: the reply was not bare JSON in our test |
| Structured output | — | ✗ Not enforced: the reply did not follow the schema in our test |
| Thinking text | — | ✗ Not returned: thinking is billed inside completion_tokens but its text is not sent back |
reasoning_effort | — | Selects the price tier (see the cost section); it is not forwarded to the model as a separate setting |
Tool setup
Base URL https://kovapi.com/v1, your KovAPI API key, and the model id gemini-3.1-pro. Request and response samples (cURL, Python, Node.js) are in the Gemini 3.1 Pro API docs.
- Open Cursor Settings → Models.
- Under API Keys, paste your KovAPI key into OpenAI API Key, turn on Override OpenAI Base URL and enter
https://kovapi.com/v1. - Add a custom model named
gemini-3.1-pro, enable it, and pick it in the chat model menu.
Differences from the official API & known limits
- The model's thinking text is not returned here, but thinking tokens are counted in
usage.completion_tokensand billed as output, as on Google's API. response_formatis not enforced here: in our weekly test a JSON-mode request came back as JSON inside a Markdown code fence, and ajson_schemarequest did not follow the schema. Strip code fences and validate the result, or use tool calling when you need a strict structure.reasoning_effortmedium,highorxhighbills the request at the second price tier shown in the cost section; omitting it or sendinglow(or any other value) uses the first tier.- Streaming responses arrive in one burst once the model has finished thinking, so the time to first token is close to the full response time.
- Each request's
usage.prompt_tokensincludes a fixed overhead of roughly 250 tokens (measured with short prompts), billed at the input rate. - Occasionally a call is served without forwarding
tools,response_format,stop,top_p,nandseed; such a call accepts onlydata:image URLs. - A prompt that the model provider blocks returns an error and is not billed.
- Claude Code cannot use this model (it speaks only the Anthropic Messages API, which KovAPI serves for Claude models). Use it from Cursor, Cline, n8n, Open WebUI or any OpenAI SDK.
Gemini 3.1 Pro API: frequently asked questions
How do I call Gemini 3.1 Pro with the OpenAI SDK?
https://kovapi.com/v1, use your KovAPI key, and set model to gemini-3.1-pro. No Google SDK or Google Cloud project is needed.How much does Gemini 3.1 Pro cost on KovAPI compared with Google?
reasoning_effort is omitted, at any prompt length. Google lists $2 / $12 per 1M for prompts up to 200K tokens and $4 / $18 above that.Are failed requests billed?
Live Test · Playground
Try out Gemini 3.1 Pro right here (available after login).
Input
Advanced
Conversation
About Gemini 3.1 Pro
Gemini 3.1 Pro is Google's next-generation flagship reasoning model, built for complex multi-step reasoning, long-context comprehension, and multimodal inputs such as text and images. On KovAPI it is served through an OpenAI-compatible API: set the model to "gemini-3.1-pro" and point your base_url at KovAPI to reuse the OpenAI SDK directly. It defaults to the fast low-effort tier; pass reasoning_effort="high" to switch to the deep-reasoning tier (billed at the Preview rate). Billing is pay-as-you-go, and failed requests are not charged.
Use cases
Summarize, extract from, and answer questions across contracts, reports, and papers, using the long context window to hold an entire document while keeping the reasoning coherent.
Tackle multi-step reasoning, math and logic problems, task decomposition, and solution planning where a clear, structured chain of analysis matters.
Combine text with image inputs for chart interpretation, screenshot Q&A, and visual description, all handled through a single chat interface.
Generate, explain, and refactor code, leaning on the long context to read whole files or modules and assist with debugging and design review.
Act as the core model in retrieval-augmented generation, feeding retrieved passages alongside the user question to produce grounded, citable answers.
How to choose
Pick the flagship Gemini 3.1 Pro when you need the deepest reasoning, the longest context, and multimodal support; if low latency and cost-efficiency for everyday chat matter more, reach for Google's faster siblings Gemini 3 Flash, Gemini 3.5 Flash, or Gemini 2.5 Flash. To compare flagships across providers, switch to GPT-5.5 or Claude Sonnet 4.6 the same way: they all live behind the one OpenAI-compatible API, so only the model field changes.
Related Models
Explore other models you can integrate.
