Google has released Gemini 3.1 Pro, an updated version of its flagship AI model, emphasizing improvements in problem-solving and reasoning. The model is available in preview for developers and consumers starting today. It builds on the Gemini 3 release from November.
Google introduced Gemini 3.1 Pro on February 19, 2026, following the launch of Gemini 3 in November. The company describes the new model as equipped to handle "your hardest challenges," with enhancements in reasoning and complex problem-solving. This update powered recent improvements to Google's Deep Think tool, announced the previous week.
Benchmark results highlight the model's advances. On Humanity’s Last Exam, which assesses advanced domain-specific knowledge, Gemini 3.1 Pro achieved 44.4 percent, surpassing Gemini 3 Pro's 37.5 percent and OpenAI’s GPT 5.2's 34.5 percent. In the ARC-AGI-2 evaluation, featuring novel logic problems, the score rose to 77.1 percent from Gemini 3's 31.1 percent, approaching competitors' scores in the 50s and 60s.
However, on the Arena leaderboard, Gemini 3.1 Pro trails in some areas. For text generation, Claude Opus 4.6 leads with 1504 points, ahead by four points. In code tasks, it lags behind Opus 4.6, Opus 4.5, and GPT 5.2 High. Google showcased the model's strengths in generating graphics and simulations, such as more elegant SVGs, and noted better performance for abstract questions and agentic workflows, where it nearly doubled its APEX-Agents benchmark score.
Availability includes preview access in AI Studio and Antigravity IDE today, with enterprise rollout via Vertex AI and Gemini Enterprise. Regular users can access it through the Gemini app and NotebookLM. Pricing remains $2 per million input tokens and $12 per million output tokens, with a 1 million input token and 64,000 output token context window. Google anticipates a similar update for its Flash model soon.
