Google DeepMind released Gemini 3.1 Pro on February 19, 2026, and the numbers are staggering. With 77.1% on ARC-AGI-2 (more than double Gemini 3 Pro), 80.6% on SWE-Bench, and zero price increase, this is the best value proposition in AI right now.
Performance That Speaks for Itself
- ARC-AGI-2: 77.1% (vs 31.1% for Gemini 3 Pro)
- SWE-Bench Verified: 80.6%
- GPQA Diamond: 94.3%
- Context Window: 1 million tokens
Three-Tier Thinking System
The biggest change is the three-tier thinking system. Previous versions had binary low and high modes. The new Medium parameter lets developers balance output latency against reasoning depth. Huge for production apps needing fast responses for simple queries but deep reasoning for complex ones.
Pricing: The Best Part
$2 per million input tokens, $12 per million output tokens – same as Gemini 3 Pro. A free upgrade with 2.5x the reasoning capability.
Known Issues
Latency spikes during peak demand. Some tests showed up to 104 seconds for basic inputs.
Our Verdict
Best value in AI right now. The ARC-AGI-2 score proves genuine reasoning, not just pattern matching. At the same price as before, zero reason not to upgrade.
Rating: 9/10
