Google has launched Gemini 3.1 Pro Preview, an upgraded version of its flagship AI model focused on advanced reasoning. The rollout started on February 19, 2026, and is available via the Gemini app, NotebookLM, Gemini API, Vertex AI, Gemini Enterprise, Android Studio, Gemini CLI, and Google Antigravity.
Gemini 3.1 Pro is designed for complex, multi-step tasks, offering refined performance, higher reliability, and improved agentic workflows. Independent benchmarks highlight its leap in intelligence: a 77.1% score on ARC-AGI-2, more than doubling Gemini 3 Pro’s 31.1%, and top scores on APEX-Agents for multi-step tasks in Google Workspace.
The model also leads on the Artificial Analysis Intelligence Index with a score of 57, outperforming rivals like Claude Opus 4.6. It excels in physics reasoning, scientific coding, knowledge accuracy, and agentic task handling, all while running at roughly half the inference cost of some competitors.
Early demos show Gemini 3.1 Pro’s creative strengths, including code-generated animated SVGs of whimsical animals performing stunts—sharp, scalable, and tiny in file size.
Some users experienced launch-day issues, like missing CLI access, bugs in Google Antigravity, and occasional hallucinations. Google has quickly addressed many of these, and community feedback remains positive.
Pricing remains consistent with Gemini 3 Pro: $2 per million input tokens and $12 per million output tokens under 200,000 tokens, with higher rates for longer contexts.
Gemini 3.1 Pro targets reasoning-heavy domains like coding, scientific research, enterprise automation, and creative prototyping, giving developers and knowledge workers new ways to tackle complex problems in 2026.
Access is expanding, with full stabilization expected soon.
Follow on Google News
