Gemini 3.1 Deep Think

What is Gemini 3.1 Deep Think?

Deep Think is a reasoning mode rather than a separate model: the highest thinking level of Gemini 3.1 Pro in the Gemini app. It works through several lines of reasoning in parallel before it answers, so a reply usually takes a few minutes, and Gemini tells you when it is ready. Google built this version with scientists for research problems where the data is messy and there may be no single right answer.

It is Google's current Deep Think and nothing has replaced it, but Google labels it experimental and may pause or remove it without notice. It needs Google AI Ultra; API access is limited to an early access programme for selected researchers, engineers and companies.

What changed from the first Gemini 3 Deep Think?

It moved to Gemini 3.1 Pro, was retuned with scientists for research and engineering, and scores far higher on Google's tests.

Base model

It now runs on Gemini 3.1 Pro instead of Gemini 3 Pro.

Scores

Humanity's Last Exam without tools rose from 41.0% to 48.4%, and ARC-AGI-2 from 45.1% to 84.6%, both verified by the ARC Prize Foundation.

Science focus

Google added physics and chemistry Olympiad results and aimed it at messy research and engineering data.

API

Selected researchers, engineers and companies can now apply to test it through the Gemini API.

Who is Gemini 3.1 Deep Think best for?

Gemini 3.1 Deep Think is best for researchers and engineers with hard maths, science or design problems who can wait a few minutes for an answer and already pay for Google AI Ultra.

• Hard maths and proofs

  It reached gold-medal level on IMO 2025 problems in Google's tests, and a Rutgers mathematician used it to find a flaw in a paper that had passed peer review.

• Physics and chemistry research

  Google's own tests put it at 87.7% on the International Physics Olympiad theory paper and 82.8% on the Chemistry Olympiad theory paper.

• Engineering design work

  Google aims it at prototyping parts, checking experimental data and tuning complex system designs, not everyday chat.

• Google AI Ultra subscribers

  It costs nothing extra on the plan, so hard questions can go to it while everyday ones stay on the faster thinking levels.

What's included with Gemini 3.1 Deep Think?

Parallel reasoning, answers that finish in the background, a 1M-token context, and access on the web and in the phone apps.

Parallel reasoning

It explores several hypotheses at once and refines them over rounds before it commits to an answer.

Answers in the background

You can leave the chat while it works; Gemini marks the finished reply on the web or sends a phone notification.

1M-token context

Ultra gives it room for long papers, datasets or codebases in a single conversation.

Sketch to 3D print

Google's launch demo shows it turning a drawing into a 3D-printable file by modelling the shape.

Web and mobile

It is a thinking level of the Pro model in the Gemini web app and the Android and iOS apps.

How does Gemini 3.1 Deep Think work?

1. Pick the Pro model

   In the Gemini app on the web or your phone, open the model menu and choose Pro.

2. Set the thinking level

   Open the menu again and choose Deep Think under Thinking level.

3. Send the problem and wait

   Answers usually take a few minutes; start another chat meanwhile and Gemini alerts you when the reply is ready.

How much does Gemini 3.1 Deep Think cost?

Gemini 3.1 Deep Think comes only with Google AI Ultra, which costs $99.99 a month for 5x the usage limits of Google AI Pro or $199.99 a month for 20x.

Deep Think is not part of the Free plan, Google AI Plus ($4.99 a month) or Google AI Pro ($19.99 a month); it is also offered with a Google AI Ultra for Business licence. Deep Think answers use more of your allowance, which resets every 5 hours up to a weekly cap, and you can buy AI credits to extend it. Google calls Deep Think experimental and may pause or remove it. These are US prices; the page lists no yearly prices, and plans can be cancelled at any time without refunds for part of a billing period. Full pricing on gemini.google. We checked these prices on 28 Sep 2026.

How do readers rate Gemini 3.1 Deep Think?

No one has rated Gemini 3.1 Deep Think yet. Be the first.

Gemini 3.1 Deep Think pros and cons

It posts Google's highest reasoning scores and users praise its best answers, but it is locked to Google AI Ultra, slow and rationed, and recent users say its quality has slipped.

What are the best Gemini 3.1 Deep Think alternatives?

Six tools on Recaply do Gemini 3.1 Deep Think's job. DeepSeek-V3.2-Speciale, Nomos 1 and NousCoder-14B are open source, so you can run them on your own server.

Gemini 3.1 ProFree

Choose it when you want answers in seconds or need an API: it is the model Deep Think runs on, with higher limits on Google AI Pro at $19.99 a month and $2 / $12 per million tokens on the Gemini API.

GPT-6 AstraPaid

Choose it when your hard problems need an API and tools: it costs $10 / $50 per million tokens on the OpenAI API and comes with ChatGPT Plus at $20 a month, though Plus allows only an estimated 5 to 45 Astra messages every five hours.

Claude Fable 5.1Paid

Choose it when the hard work is long coding or agent runs rather than single maths or science questions; Claude Max includes it from $100 a month, and the API charges $10 / $50 per million tokens.

DeepSeek-V3.2-SpecialeOpen Source

Choose it when you want open weights for hard maths and coding problems without a Google AI Ultra subscription.

Nomos 1Open Source

Choose it when you want an Apache-licensed maths reasoning model you can download, not a Google AI Ultra plan.

NousCoder-14BOpen Source

Choose it when you want a 14B Apache-licensed model for competitive programming that you can run locally.

Questions about Gemini 3.1 Deep Think