Gemini 3.1 Pro: Features, Pricing, and Coding Use
Gemini 3.1 Pro is Google's high-capability multimodal AI model for reasoning, coding, and tool-driven work. The current API version is Preview, exposed through the model identifier gemini-3.1-pro-preview; it is not a stable API release. Preview availability, limits, and model behavior can change before a stable version becomes available.
The model was released on February 19, 2026. It accepts text, images, video, audio, and PDF input and produces text output, with a documented window of up to 1,048,576 input tokens and 65,536 output tokens.
Gemini 3.1 Pro vs 3 Pro: What Changed
Gemini 3.1 Pro builds on Gemini 3 Pro with improvements aimed at complex reasoning and agentic workflows. Its focus includes software engineering, multimodal understanding, and tasks that combine planning with tools. A migration trial can keep the prompt and tool schema fixed while comparing quality and latency.
| Migration dimension | Trial detail |
|---|---|
| Model identity | Exact Preview model string |
| Thinking | Thinking level or budget used |
| Tools | Function calls, code execution, and grounding behavior |
| Context | Input size, output size, and any cached tokens |
New Capabilities & Improvements
The API supports thinking controls, function calling, code execution, structured output, search grounding, Maps grounding, and URL context. A custom-tools endpoint is also available for tool-heavy agent workflows.
Choose capabilities from the task rather than enabling every tool. For example:
- “Inspect this PDF specification and repository, then map each requirement to an implementation file.”
- “Use the provided tools to reproduce the bug, propose a repair, and stop before modifying data.”
- “Return a structured migration plan with owners, risks, tests, and rollback steps.”
Coding Benchmarks and Evaluation
Published evaluation results for Gemini 3.1 Pro depend on harness, prompting, tools, and scoring. Use them to choose a candidate for testing, not to predict a repository outcome.
A practical coding trial should include one bounded bug, one multi-file feature, and one review task. Score functional correctness, unnecessary edits, tool-call validity, test evidence, latency, and reviewer corrections.
Pricing
For prompts up to 200K tokens, the API pricing is $2 per million input tokens and $12 per million output tokens. Above 200K tokens, the listed rates are $4 input and $18 output per million tokens. API use has no free tier for this model, although Google AI Studio may allow free experimentation under its own terms.
Long-context requests can cross the pricing threshold quickly. Estimate input, cached context, tool traffic, and expected output before selecting the model for a repetitive agent task.
Gemini 3.1 Pro in Verdent
Verdent's built-in catalog lists gemini-3.1-pro-preview at the short-context $2 input and $12 output rates per million tokens. The Verdent model-pricing page also helps separate that tier from long-context requests. Keep the Preview identifier with the task results so a later model change does not blur the comparison.
Migration and Model Selection
Upgrade when your current model struggles with multimodal evidence, long inputs, complex tool plans, or difficult coding tasks, and a controlled test shows lower review effort. Stay with a faster or lower-cost tier when the task is short, high-volume, and already reliable.
Run the same prompts against your current baseline. A useful upgrade decision combines quality, latency, cost, and the amount of human correction required.
Run a Preview Migration Trial
Use one representative task, preserve the same tool schema, and compare total review effort before changing the project baseline.
Frequently Asked Questions
Is Gemini 3.1 Pro a Preview model?
Yes. Its current API identifier is gemini-3.1-pro-preview, so teams should monitor availability and behavior changes.
When was Gemini 3.1 Pro released?
Gemini 3.1 Pro was announced on February 19, 2026. The May 2026 date sometimes associated with it is not the model's release date.
How much does the Gemini 3.1 Pro API cost?
Up to 200K input tokens, the listed rates are $2 per million input tokens and $12 per million output tokens. Longer prompts use higher rates.
Can I use Gemini 3.1 Pro in Verdent?
Yes. Verdent lists the Preview model in its built-in catalog. The higher long-context price tier still applies when prompts exceed 200K tokens.