Google DeepMind · released
Gemini 3.1 Pro
The best published GPQA score, and near-saturation on ARC-AGI-1.
Overview
Google's reasoning flagship. It holds the highest GPQA Diamond figure we could source, and effectively saturates ARC-AGI-1 at 98%. Pricing is tiered by context length rather than flat, which makes it cheap on short prompts and less so on the long ones it is built for.
Solthar DNA rating, out of 5
- reasoning: 5
- coding: 4
- writing: 4
- vision: 5
- speed: 4.5
- value: 4
- context: 5
- tools: 4
Verdict
Strongest science and multimodal reasoning, and the most generous free access of the frontier models.