Updated

A large faceted crystal prism splitting light into fine beams on a dark background, abstract hero artwork for Gemini 3.1 Pro Preview

Google

Gemini 3.1 Pro Preview

Google's deep-reasoning Pro model, still a preview in Google's API, for agentic coding.

Chat with Gemini 3.1 Pro Preview in Super AgentView pricing

What is Gemini 3.1 Pro Preview?

Gemini 3.1 Pro is Google's Pro reasoning model, released on February 19, 2026 and still offered only as a preview (gemini-3.1-pro-preview) in Google's API, with no stable version yet.

Google calls it a step forward in core reasoning, with a verified 77.1% on ARC-AGI-2, more than double Gemini 3 Pro's score. It is built for tasks where a simple answer isn't enough, such as visual explanations, data synthesis and creative coding like animated SVGs, and for software engineering and agent workflows that need precise tool use. Its 1M-token context window accepts text, images, video, audio and PDFs.

Provider
Google
Context window
1M tokens
Input
Text, Image, Video, Audio, PDF
Reasoning
Low, medium, high
Max output
65,536 tokens
Knowledge cutoff
January 2025
Released
Feb 19, 2026

How Gemini 3.1 Pro Preview performs

Published by Google

Results Google reported for Gemini 3.1 Pro Preview. Each lab runs its own setup, so compare these with other models only inside the same table.

Benchmark results Google published for Gemini 3.1 Pro Preview
BenchmarkScoreSetting
Humanity's Last Exam44.4%no tools, high thinking
ARC-AGI-277.1%ARC Prize Verified, high thinking
GPQA Diamond94.3%no tools, high thinking
Terminal-Bench 2.068.5%Terminus-2 harness, high thinking
SWE-Bench Verified80.6%single attempt, high thinking
SWE-Bench Pro (Public)54.2%single attempt, high thinking
LiveCodeBench Pro2,887 Elohigh thinking
BrowseComp85.9%search, Python and browsing, high thinking
MMMU-Pro80.5%no tools, high thinking
MCP Atlas69.2%high thinking

Source: deepmind.google

What you can do with Gemini 3.1 Pro Preview

  • Verified 77.1% on ARC-AGI-2, more than double Gemini 3 Pro

  • A more grounded, factually consistent experience than the Gemini 3 Pro series it refines, per Google

  • Creative coding such as animated SVGs

  • Works across text, images, video, audio and PDFs

  • 1M tokens of context for long multimodal briefs

How to prompt Gemini 3.1 Pro Preview

  • Ask it to explain a concept with an animated SVG or chart code; Google built it for answers that go past plain text.
  • Use medium effort as a middle ground between speed and depth.
  • Attach mixed media (a PDF, a clip, a voice note) in one message when they belong together.

Where Gemini 3.1 Pro Preview fits best

  • Visual explanations and data synthesis

  • Animated SVGs and other creative code

  • Long multimodal briefs

  • Agent runs that need precise tool use

How Gemini 3.1 Pro Preview compares

Gemini 3.1 Pro Preview compared with other chat models on Flashloop
ModelLMArena textLMArena codingContext windowReasoning
Gemini 3.1 Pro Preview1487#191522#321M tokensLow, medium, high
Kimi K31487#181539#111M tokensOff, low, medium, high, max
GPT 5.6 Sol1485#211534#171.05M tokensOff to max
Gemini 3.7 Flash1490#161520#361M tokensLow, medium, high
Claude Opus 51491#131528#231M tokensOff to max

Good to know before you start

  • Still a preview model in Google's API, with no stable release
  • Low, medium and high effort only; reasoning can't be turned off or set to max

Start chatting with Gemini 3.1 Pro Preview

Chat with Gemini 3.1 Pro Preview in Super Agent

  1. Select Gemini 3.1 Pro Preview

    Open Super Agent on Flashloop and choose Gemini 3.1 Pro Preview in the model picker.

  2. Enter your prompt

    Ask for what you need: an answer, a plan, or the images and videos you want the agent to make.

  3. Send and keep going

    Send the message. Gemini 3.1 Pro Preview answers in Super Agent and runs any generations you asked for.

Questions about Gemini 3.1 Pro Preview

Is Gemini 3.1 Pro a preview model?

Yes. Google still serves it as gemini-3.1-pro-preview, and there is no stable Gemini 3.1 Pro ID. Google hasn't announced a shutdown date for the preview.

What is Gemini 3.1 Pro Preview good at?

Hard reasoning, agentic coding with precise tool use, and answers that need more than text, such as visual explanations, data synthesis and animated SVGs.

Which reasoning levels does Gemini 3.1 Pro Preview support in Super Agent?

Low, medium and high. Super Agent starts it at low.

How does Gemini 3.1 Pro Preview differ from the Flash models?

It is Google's deeper-reasoning Pro tier. The Flash models are built for speed and efficiency.

More chat models on Flashloop

Browse all models