Gemini 3.1 Pro

O Google DeepMind

Założona 2010 · Mountain View, USA / London, UK

Google DeepMind is the merged AI research organisation formed in April 2023 by combining Google Brain (founded inside Google in 2011) with DeepMind (founded in London in 2010 by Demis Hassabis, Shane Legg and Mustafa Suleyman, acquired by Google in 2014). Demis Hassabis leads the unit as CEO. DeepMind authored seminal papers including 'Attention Is All You Need' (Google Brain, 2017), AlphaGo (2016), AlphaFold (2018-2021, awarded the 2024 Nobel Prize in Chemistry), AlphaZero, Chinchilla scaling laws and the Gemini Technical Report. The Gemini family launched in December 2023 (Ultra, Pro, Nano), followed by Gemini 1.5 Pro with 1M+ context (early 2024), Gemini 2.0 Flash (December 2024), Gemini 2.5 Pro with Deep Think (March/May 2025), Gemini 3 Pro (late 2025), Gemini 3.1 Pro (February 2026) and Gemini 3 Flash (April 2026). Google DeepMind also ships Imagen, Veo, Lyria and NotebookLM and powers AI features across Google Search, Workspace and Android.

Architektura

Sparse Mixture-of-Experts Transformer (natively multimodal, Deep Think reasoning)

Gemini 3.1 Pro was released February 19, 2026 as Google DeepMind's flagship frontier model. It is a natively multimodal Sparse Mixture-of-Experts Transformer that ingests text, image, audio and video tokens through a shared embedding space, building directly on the Gemini 3 Pro architecture (late 2025) and Gemini 2.5 Pro (early 2025). The input context window is 1,048,576 tokens, the same size as Gemini 2.5 Pro. Pretraining used Google's TPU v6e infrastructure on a multi-trillion-token corpus mixing web text, code, books, scientific papers, image-text pairs, audio waveforms and video frames, with a knowledge cutoff in late 2025. Post-training combined supervised fine-tuning, RLHF, reinforcement learning against verifiable rewards on math and code, and a refreshed 'Deep Think' reasoning stage that teaches the model to allocate test-time thinking budgets and emit long internal chains-of-thought before its final answer. Tool use, function calling, structured output, Search grounding and the Code Execution tool are first-class. Safety training followed Google's Frontier Safety Framework v2.

Parametry

Undisclosed (sparse MoE, total parameters in the hundreds of billions, active per-token undisclosed)

Kontekst

2 000 000 tokenów

Możliwości

• 1,048,576-token input context window with native long-video and long-audio support
• Built-in Deep Think reasoning mode with adjustable thinking budget
• Natively multimodal: text, image, audio and video in a single pass
• Top scores on GPQA Diamond, AIME 2026 and Humanity's Last Exam
• Strong coding performance on SWE-bench Verified and LiveCodeBench
• Function calling, JSON schema and parallel tool calls
• Search grounding and Code Execution tools built into the API
• Cross-lingual reasoning across 100+ languages
• Available via Vertex AI, AI Studio and the Gemini app (paid-only since April 2026)
• 2x batch pricing reduction for large jobs
• Best for: long-document and long-video analysis, scientific reasoning, agentic workflows, complex multimodal extraction.

Trening i licencja

Pretrained on a multi-trillion-token mixture of web text, code, books, scientific papers, licensed third-party text, audio waveforms, image-text pairs and video frames. Knowledge cutoff in late 2025. Post-training uses supervised fine-tuning, RLHF, RL against verifiable rewards and Deep Think reasoning training.

Licencja: Proprietary commercial license via Google AI Studio, Vertex AI and the Gemini app. Paid-only since April 1, 2026.

Testy bezpieczeństwa: Evaluated under Google DeepMind's Frontier Safety Framework v2 covering CBRN, cyber, persuasion and ML R&D risks, with internal red teams and external evaluators.