Gemini 3.1 Pro
Google DeepMind's multimodal reasoning model, the next step after Gemini 3 Pro
Gemini 3.1 Pro is a natively multimodal reasoning model from Google DeepMind, described in a model card published 19 February 2026 as the company's most advanced model for complex tasks at that date. It builds on Gemini 3 Pro and handles text, audio, images, video and whole code repositories with a context window of up to one million tokens.
- Published 19 February 2026 as Google's most advanced model for complex tasks
- Context window of up to 1M tokens across text, images, audio and video
- 77.1% on ARC-AGI-2 abstract reasoning puzzles, versus 31.1% for Gemini 3 Pro
- 94.3% on GPQA Diamond scientific knowledge without tools
- Based directly on Gemini 3 Pro, with architecture details deferred to that model card
What it is
Gemini 3.1 Pro is a large language model from Google DeepMind, presented as the next iteration in the Gemini 3 series. The company describes it as a natively multimodal reasoning model and, at the date of its model card, its most capable offering for difficult work. It accepts text, images, audio and video, and can work through large collections of material including complete code repositories. The context window runs to one million tokens. Google DeepMind positions it for tasks needing long context, multimodal understanding, strategic planning and step-by-step problem solving.
How it came about
The model card for Gemini 3.1 Pro was published on 19 February 2026. The model follows Gemini 3 Pro, on which it is directly based; the card points readers to that earlier model for details of architecture, training data, hardware and software. Google DeepMind notes that model cards may be revised over time as models are improved or updated evaluations become available.
Inside the model
Google DeepMind gives no separate architecture or training description for Gemini 3.1 Pro, referring instead to the Gemini 3 Pro model card for architecture, training data, data processing, hardware and software. What is stated is that the model is natively multimodal and reasoning-focused, taking text, images, audio and video with a context window of up to one million tokens. The company says it can handle entire code repositories alongside other large, mixed-format inputs.
What it can do
Reported results from February 2026 show gains over Gemini 3 Pro on many reasoning and multimodal tests. With high thinking settings, it scored 44.4% on Humanity's Last Exam without tools and 51.4% with search and code, 77.1% on ARC-AGI-2 puzzles, 94.3% on GPQA Diamond, 80.6% on SWE-Bench Verified and an Elo of 2887 on LiveCodeBench Pro. Multilingual Q&A reached 92.6% on MMMLU. On long-context MRCR v2 at 128k it scored 84.9%, but only 26.3% at the one million token pointwise setting. Some rivals scored higher on individual tests, such as GDPval-AA.
How to use it
The model reaches users through an application program interface rather than downloadable weights. Google DeepMind says downstream providers can serve it via API under the relevant terms of use, with no special hardware or software required. Access routes named are AI Studio and the Gemini API, governed by the Gemini API Additional Terms of Service, and Vertex AI, under the Google Cloud Platform Terms of Service.
Safety and testing
Internal automated safety evaluations, run without human review or red teaming, showed small changes against Gemini 3 Pro: text-to-text safety up 0.10%, multilingual safety up 0.11%, image-to-text safety down 0.33%, tone up 0.02% and unjustified refusals down 0.08%. Manual review of flagged cases found the losses were mostly false positives or not serious. Specialist red teams outside the development group tested the model, and child safety launch thresholds were met. Google DeepMind's Frontier Safety Framework covers CBRN, cyber, harmful manipulation, machine learning R&D and misalignment risks.
Questions people ask about Gemini 3.1 Pro
What is Gemini 3.1 Pro?
It is a natively multimodal reasoning model from Google DeepMind, part of the Gemini 3 series and built on Gemini 3 Pro. It handles text, images, audio, video and whole code repositories, with a context window of up to one million tokens.
How is it different from Gemini 3 Pro?
Google DeepMind reports it significantly outperforms Gemini 3 Pro on reasoning and multimodal benchmarks. Examples from February 2026 include ARC-AGI-2 at 77.1% versus 31.1%, and Humanity's Last Exam at 44.4% versus 37.5% without tools.
Can I download the model?
No downloadable weights are offered. Google DeepMind distributes it through an API, available via AI Studio and the Gemini API under the Gemini API Additional Terms of Service, or through Vertex AI under Google Cloud Platform terms.
What can it take as input?
Text strings such as prompts and documents, plus images, audio and video files. The context window reaches up to one million tokens, and the company says it can work through entire code repositories and other large mixed-format datasets.
How does it handle safety?
Automated internal evaluations showed small shifts against Gemini 3 Pro on content safety, tone and refusals, with manual review of flagged cases. Specialist red teams outside the model team tested it, and it met launch thresholds for child safety.
Good reading and sources
There is no English Wikipedia article on Gemini 3.1 Pro yet, so the facts on this page come from the sources listed above: deepmind.google/models/model-cards/gemini-3-1-pro/. Some are the subject's own pages, so its claims about itself are reported as its claims. EveryLLM wrote the text in its own words, with AI assistance, and compares it with the sources to make sure no passage is copied.
It may contain mistakes, and figures change over time. Nothing here is investment or buying advice. If something is wrong, tell us. Last refreshed 2026-11-14.