Launches Google

Google Unveils Gemini 3.1 Pro with Massive Reasoning Leap

Google announces Gemini 3.1 Pro, more than doubling reasoning performance on ARC-AGI-2. The first '.1' increment signals accelerated AI development.

Published February 21, 2026
Visit Source
gemini google ai-model reasoning ai-launch

Google’s Fastest Model Iteration Yet

Google announced Gemini 3.1 Pro on February 19, 2026 — its newest and most capable reasoning model. The release marks the first time Google has used a “.1” version increment, signaling an accelerated pace of model development that puts it in direct competition with Anthropic’s recent Claude 4.6 releases.

The headline number: Gemini 3.1 Pro scored 77.1% on ARC-AGI-2, more than doubling the performance of standard Gemini 3 Pro on the same benchmark. ARC-AGI-2 tests abstract reasoning and novel problem-solving — areas that have traditionally been challenging for large language models.

What Makes 3.1 Pro Different

Gemini 3.1 Pro is specifically optimized for agentic workflows and deep reasoning tasks. It excels at multi-step planning, code generation, complex analysis, and tasks that require maintaining coherence across long reasoning chains.

The model is available for free preview in Google AI Studio and Vertex AI, making it immediately accessible to developers. It also supports Gemini CLI and Android Studio integration, positioning it as a direct tool for software development workflows.

A Busy February for Google AI

The 3.1 Pro announcement caps an extraordinarily active month for Google’s AI division:

  • Gemini 3 Flash became the default model across the Gemini app and AI Mode in Search, beating its predecessor on 18 of 20 benchmarks while being 3x faster
  • Gemini 3 Deep Think received a major upgrade on February 12, with improved practical applications exclusive to AI Ultra subscribers
  • Google rebranded its subscription tiers from “Gemini Advanced” to “Google AI Pro” ($19.99/month) and “Google AI Ultra” ($249.99/month)
  • Personal Intelligence launched, connecting Gemini to users’ Gmail, Photos, YouTube, and Search data for personalized responses

The AI Reasoning Race Heats Up

Gemini 3.1 Pro enters a crowded field. Anthropic’s Claude Opus 4.6 (released February 5) leads SWE-bench Verified at 80.8%, while OpenAI’s o3 and GPT-5 series continue to push boundaries in mathematical and scientific reasoning.

Google’s strategy appears focused on rapid iteration — getting models from research to production faster than ever. The “.1” naming convention suggests we may see 3.2 and beyond before a full Gemini 4 generation, similar to how Anthropic has iterated within the Claude 4.x family.

What This Means for Developers

Gemini 3.1 Pro is available now in preview via Google AI Studio with no cost for experimentation. Developers building agentic applications — autonomous coding tools, research agents, multi-step workflow automation — should evaluate it against current options. The model’s strength in abstract reasoning and planning makes it particularly suited for agent frameworks that need reliable multi-step execution.

Production pricing and general availability details are expected in the coming weeks.