Gemini 3.1 Pro

What is Gemini 3.1 Pro?

Gemini 3.1 Pro is the upgraded core of Google's Gemini 3 Pro line. It reads text, images, video, audio and PDFs across a 1M-token window, writes up to 64K tokens of text, and thinks at low, medium or high depth before it answers. Its knowledge stops at January 2025, so turn on Google Search grounding for anything newer.

It is still a preview model with no shutdown date, and Google lists Gemini 3.5 Pro as coming soon. Meanwhile, Google's own tests show newer Flash models beating it on agentic coding at a lower price, so it now fits reasoning-heavy work better than long agent runs.

What changed from Gemini 3 Pro?

Much stronger reasoning and agentic coding, a new medium thinking level and a custom-tools endpoint; Gemini 3 Pro itself has been shut down.

Reasoning

77.1% on ARC-AGI-2, up from 31.1% for Gemini 3 Pro, and 44.4% on Humanity's Last Exam without tools, up from 37.5%.

Agentic coding

68.5% on Terminal-Bench 2.0 and 80.6% on SWE-Bench Verified, up from 56.9% and 76.2%.

Thinking

Adds a medium thinking level between low and high, for a middle ground on cost and speed.

Custom tools

A second endpoint, gemini-3.1-pro-preview-customtools, for agents that mix bash with their own tools.

Gemini 3 Pro

Shut down on the Gemini API; requests to its old model ID now go to 3.1 Pro.

Who is Gemini 3.1 Pro best for?

Gemini 3.1 Pro is best for hard reasoning, code-built visuals, long documents and video, and research or translation work, more than for long agentic coding runs.

• Hard reasoning problems

  It scored 77.1% on ARC-AGI-2 and 94.3% on GPQA Diamond in Google's tests, the kind of new-logic and science questions where a quick answer falls short.

• Code-built visuals

  It writes animated SVGs, live data dashboards and interactive 3D scenes from a text prompt, and users single out its SVG and 3D output.

• Long documents, video and codebases

  A 1M-token input window takes text, images, video, audio and PDFs, up to about 45 minutes of video with sound in one prompt on Google Cloud.

• Research and translation

  Users single it out for research, knowledge questions and literary translation, where coding benchmarks matter less.

What's included with Gemini 3.1 Pro?

Three thinking levels, a custom-tools endpoint, Google Search and Maps grounding, long video and audio input, cheaper Batch and Flex tiers, and reserved capacity on Google Cloud.

Three thinking levels

Low, medium and high (the default) trade answer depth against speed and cost; there is no minimal level for the fastest replies.

Custom tools endpoint

A separate gemini-3.1-pro-preview-customtools endpoint favours your own tools, such as view_file or search_code, over bash commands.

Grounding and built-in tools

Google Search and Maps grounding, code execution, URL context and function calling, which can be combined with structured outputs.

Long media input

On Google Cloud, one prompt can hold about 45 minutes of video with audio, 8.4 hours of audio or 3,000 images.

Batch, Flex and Priority tiers

Half-price Batch and Flex for work that can wait, and a Priority tier that is served ahead of other traffic.

Provisioned throughput

Companies can reserve capacity for the main endpoint on Google Cloud's Gemini Enterprise Agent Platform.

How much does Gemini 3.1 Pro cost?

Gemini 3.1 Pro costs $2 per million input tokens and $12 per million output tokens on the Gemini API for prompts up to 200K tokens, or half that on Batch and Flex; there is no free API tier, but the free Gemini app gives varying access.

API prices are for the Gemini Developer API in US dollars. Priority costs $3.60 / $21.60 per million tokens ($7.20 / $32.40 over 200K tokens). Paid API use needs a linked billing account and at least $5 of prepaid credit, which expires after 12 months; paid-tier prompts are not used to improve Google's products. Google Search grounding includes 5,000 free requests a month shared across Gemini 3 models, then $14 per 1,000. In the Gemini app, Google AI Plus costs $4.99 a month and Google AI Ultra from $99.99 a month, at US prices. Full pricing on ai.google.dev. We checked these prices on 28 Sep 2026.

How do readers rate Gemini 3.1 Pro?

No one has rated Gemini 3.1 Pro yet. Be the first.

Gemini 3.1 Pro pros and cons

It posts top reasoning scores at half Claude Opus 5.5's input price, but it is still a preview, thinks slowly on hard prompts, and users rate it below Claude for agentic coding.

What are the best Gemini 3.1 Pro alternatives?

Six tools on Recaply do Gemini 3.1 Pro's job. DeepSeek-V3.2, DeepSeek V4 Pro and Ring-2.6-1T are open source, so you can run them on your own server.

Gemini 3.8 FlashFree

Choose it when you want Google's newer, generally available model for coding agents: it costs $1.50 / $7.50 per million tokens, has a free API tier, and its knowledge reaches March 2026 in some areas.

Claude Opus 5.5Paid

Choose it when agentic coding quality matters more than price: it leads Anthropic's Terminal-Bench 4.0 table at 66.4% and costs $4 / $20 per million tokens.

GPT-5.5Paid

Choose it when you need up to 128K tokens of output, a December 2025 knowledge cutoff or reasoning you can switch off; it costs more, at $5 / $30 per million tokens.

DeepSeek-V3.2Open Source

Choose it when you need open weights for agent work that you can self-host and fine-tune, rather than a Google-hosted model.

DeepSeek V4 ProOpen Source

Choose it when you need a frontier-size open model for coding and research that can run in your own cloud.

Ring-2.6-1TOpen Source

Choose it when you want a trillion-parameter MIT-licensed reasoning model for agent workflows on your own cluster.

Questions about Gemini 3.1 Pro