[Google DeepMind](https://deepmind.google) is accelerating. On February 19, 2026, the lab released Gemini 3.1 Pro, a major update that doubles the reasoning capabilities of its predecessor. With a score of 77.1% on ARC-AGI-2 — the benchmark that evaluates the ability to solve novel logical problems — Gemini 3.1 Pro more than doubles the performance of Gemini 3 Pro. It is proof that the AI reasoning race is not slowing down.
A leap in reasoning
The key number is 77.1% on ARC-AGI-2. This benchmark, designed to evaluate a model's ability to solve entirely novel logical problems (not simple pattern matching), is considered one of the most demanding tests in the industry. Gemini 3 Pro plateaued around 35%. Gemini 3.1 Pro more than doubles that score.
This progress is not isolated. Gemini 3.1 Pro inherits the enhanced intelligence introduced the previous week with Gemini 3 Deep Think, and extends it to a more accessible model available across all Google platforms.
Benchmarks: Gemini 3.1 Pro vs the competition
| Benchmark | Gemini 3.1 Pro | Claude Opus 4.7 | GPT-5.4 |
|---|---|---|---|
| ARC-AGI-2 | 77.1% | N/A | N/A |
| Humanity's Last Exam (no tools) | 44.4% | N/A | N/A |
| MRCR v2 8-needle (128k) | 84.9% | 84.9% (Sonnet 4.6) | 83.8% |
| Max context | 1M tokens | 1M tokens | 1.05M tokens |
| Pricing (input/output) | Variable | $5/$25 per M | $2.5/$15 per M |
Comparative performance (February-April 2026)
Vibe coding and agentic workflows
Google positions Gemini 3.1 Pro as its best model for vibe coding — the approach where you describe what you want to build in natural language and the AI handles it. The model can generate animated SVGs ready for the web, build interactive dashboards from real-time data, and transform thematic concepts into complete interfaces.
Instruction following is significantly improved, with better tool usage and enhanced agentic capabilities. Gemini 3.1 Pro is at the heart of [Google Antigravity](/blog/google-antigravity-agentic-development-platform), Google's agentic IDE where it can orchestrate complex tasks autonomously.
Availability: everywhere in the Google ecosystem
Google is deploying Gemini 3.1 Pro across a wide range of platforms:
- Gemini app — For consumers with AI Pro and Ultra plans (higher rate limits).
- NotebookLM — For research and document analysis (exclusive to Pro/Ultra).
- Google AI Studio — For developers, via the Gemini API.
- [Google Antigravity](/blog/google-antigravity-agentic-development-platform) — The agentic IDE with 2M token context.
- Vertex AI / Gemini Enterprise — For businesses.
- Gemini CLI — For developers on the command line.
- Android Studio — For mobile development.
What early users are saying
“Gemini 3.1 Pro represents a clear quality leap in our evaluations. We observed up to 15% improvement compared to the best runs of Gemini 3 Pro Preview. The model is stronger, faster, and more efficient.”
Companies including Databricks, Hostinger, and others have confirmed significant improvements, particularly in understanding 3D transformations, generating structured code, and synthesizing complex data.
Frequently asked questions
Compare AI models
Stay up to date on AI news
Get our analyses of the latest advances in artificial intelligence.
No spam. Unsubscribe in 1 click.
Google Antigravity: the agentic IDE
Google's agent-first IDE powered by Gemini 3.1 Pro.
Read the articleAI Model Comparison Guide Q2 2026
GPT-5.4, Opus 4.7, Gemini 3.1 Pro — which model should you choose?
Read the guide

