Google DeepMind launches Gemini 3.1 Pro with advanced reasoning capabilities

February 19, 2026

Details

  • Google DeepMind released Gemini 3.1 Pro, an upgraded frontier reasoning model designed for complex problem-solving workflows that require more than simple answers.
  • The model achieves more than double the performance of Gemini 3 Pro on ARC-AGI-2, a benchmark testing novel logic pattern recognition, demonstrating significant gains in reasoning ability.
  • Gemini 3.1 Pro excels at visualizing complex topics, organizing scattered data, and handling multimodal inputs (text, image, video, audio, PDF). It features a 1M-token context window and improved token efficiency.
  • The model is optimized for software engineering, agentic coding, financial modeling, and structured planning tasks, with enhanced tool orchestration and multi-step execution reliability.
  • Availability: Rolling out now in the Gemini App and exclusively for Google AI Pro and Ultra subscribers in NotebookLM. Developers can access it via preview in Google AI Studio API.

Impact

Gemini 3.1 Pro represents Google's intensifying competition with OpenAI and Anthropic in the frontier reasoning space. The more-than-doubled ARC-AGI-2 score signals meaningful progress on abstract reasoning—a critical capability gap that has historically favored competitors. The exclusive initial rollout to Google's subscription tiers (Pro/Ultra) and NotebookLM mirrors Google's strategy of driving adoption into existing products rather than competing purely on API pricing, potentially widening the moat around Google's integrated AI ecosystem. The emphasis on agentic and software engineering workflows aligns with industry momentum toward autonomous AI agents; early GitHub Copilot integration further positions Gemini 3.1 Pro against GitHub Copilot's existing model lineup, particularly OpenAI. Over the next 12 months, this may accelerate enterprise adoption if performance claims hold and latency/cost remain competitive. The release also signals Google's continued refinement of its reasoning tier strategy—introducing a "medium thinking" level alongside deeper variants—suggesting a multi-tier API pricing model to capture different cost-performance segments.

Read more