Gemini 3.1 Pro

Gemini 3.1 Pro, released on 19 February 2026, is the reasoning-focused tier of Google DeepMind's Gemini line, sitting above the speed-oriented Flash models. It accepts up to one million tokens of input — a whole codebase, roughly 8.4 hours of audio, a 900-page PDF or an hour of video in a single prompt — and returns up to 64,000 tokens. Input covers text, images, audio, video and code repositories; output is text. The jump from Gemini 3 Pro shows most clearly on tasks the model has not seen before. On ARC-AGI-2, which measures solving genuinely novel logic patterns, Google reports 77.1% — more than double the previous Pro. The model card also lists 94.3% on GPQA Diamond, 80.6% on SWE-bench Verified, 80.5% on MMMU-Pro, an Elo of 2887 on LiveCodeBench Pro and 84.9% on MRCR v2 at 128k context. Unlike Gemini 3 Deep Think, aimed at scientific and engineering research, 3.1 Pro is positioned for everyday complex reasoning across consumer, developer and enterprise use. It is available in the Gemini app, the Gemini API, Google AI Studio, Vertex AI, Google Antigravity, Gemini Enterprise and NotebookLM. Google reports the model below its alert thresholds for CBRN, harmful manipulation, ML R&D and misalignment risk, with automated safety evaluations and human red-teaming showing performance similar to Gemini 3 Pro. The first-generation Gemini 3 Pro and the later Gemini 3.6 Flash both have their own entries in this catalogue.

▸Videos

No videos yet.

▸Reviews