Best AI model for maths
Measured in Nivana’s weekly evals (updated 2026-10-25): Claude Opus scores highest on maths at 93/100. The router also weighs price and speed, so its pick depends on how hard the request is.
What the router picks
- Easy: GPT-6 Luna
- Medium: DeepSeek V4 Pro
- Hard: DeepSeek V4 Pro
Ranked by measured quality
| Model | Lab | Score | Price |
|---|---|---|---|
| Claude Opus | Anthropic | 93 | $4 in / $20 out per 1M tokens |
| DeepSeek V4 Pro | DeepSeek | 92 | $0.66 in / $1.98 out per 1M tokens |
| Gemini 4 Argon | 87 | $2 in / $10 out per 1M tokens | |
| GPT-5.5 | OpenAI | 87 | $5 in / $30 out per 1M tokens |
| Claude Sonnet | Anthropic | 84 | $2 in / $10 out per 1M tokens |
| Mistral Large | Mistral AI | 81 | $1.36 in / $4.18 out per 1M tokens |
| Gemini Flash | 80 | $0.75 in / $3.75 out per 1M tokens | |
| Kimi K3.1 Pro | Moonshot | 80 | $2 in / $10 out per 1M tokens |
| GPT-6.1 Sol | OpenAI | 77 | $2 in / $10 out per 1M tokens |
| DeepSeek Flash | DeepSeek | 77 | $0.075 in / $0.3 out per 1M tokens |
| Muse Spark | Meta | 77 | $1.25 in / $4.25 out per 1M tokens |
| Grok 4.7 | xAI | 75 | $2 in / $6 out per 1M tokens |
| Claude Haiku | Anthropic | 71 | $0.1 in / $0.5 out per 1M tokens |
| GPT-6 Luna | OpenAI | 70 | $0.1 in / $0.5 out per 1M tokens |