Best LLM for research-level math
- GPT-6 Astra leads
- 93.7%
- models ranked
- 80
FrontierMath, tiers 1 to 3 · read October 21
As of October 21, 2026, OpenAI GPT-6 Astra leads 80 models on FrontierMath, tiers 1 to 3 with 93.7%, ahead of OpenAI GPT-6.1 Sol (93.7%) and Anthropic Claude Opus 5.5 (91.2%).
The full best LLM for math ranking combines this benchmark with others.
| Rank | Model | FrontierMath, tiers 1 to 3 | Coding evidence | Coding index | Output $/Mtok |
|---|---|---|---|---|---|
| 1 | GPT-6 Astra OpenAI | 93.7% | 85.1 | $50.00 | |
| 2 | GPT-6.1 Sol OpenAI | 93.7% | 82.4 | $10.00 | |
| 3 | Claude Opus 5.5 Anthropic | 91.2% | 88.8 | $20.00 | |
| 4 | Claude Fable 5.1 Anthropic | 90.2% | 84.2 | $50.00 | |
| 5 | GPT-6 Sol OpenAI | 89.8% | 76.0 | $10.00 | |
| 6 | GPT-5.6 Sol OpenAI | 89.1% | 73.5 | $20.00 | |
| 7 | Claude Sonnet 5.5 Anthropic | 88.8% | 87.3 | $10.00 | |
| 8 | GPT-5.5 Pro OpenAI | 87.7% | Not benchmarked | $180.00 | |
| 9 | Claude Fable 5 Anthropic | 87.0% | 79.3 | $50.00 | |
| 10 | Gemini 4 Argon Google | 86.0% | 81.0 | $10.00 | |
| 11 | GPT-5.6 Terra OpenAI | 86.0% | 63.9 | $12.00 | |
| 12 | Claude Opus 5 Anthropic | 85.6% | 83.0 | $25.00 | |
| 13 | GPT-5.5 OpenAI | 85.3% | 71.3 | $30.00 | |
| 14 | GPT-5.4 Pro OpenAI | 82.5% | Not benchmarked | $180.00 | |
| 15 | GPT-5.6 Luna OpenAI | 82.1% | 58.0 | $1.20 | |
| 16 | Claude Opus 4.8 Anthropic | 80.0% | 65.0 | $25.00 | |
| 17 | GPT-6 Luna OpenAI | 78.9% | 57.3 | $0.50 | |
| 18 | GPT-5.4 OpenAI | 78.6% | 64.4 | $15.00 | |
| 19 | Claude Haiku 5.5 Anthropic | 75.1% | 70.7 | $0.50 | |
| 20 | Qwen3.8 Max (0902) Qwen | 74.7% | 62.4 | $6.00 |
How this list is built
Generated from the dataset on every refresh. Nobody picks the order by hand.
Models by their FrontierMath, tiers 1 to 3 result, as the board publishes it, best first. Unpublished research-level math problems written by mathematicians, run by Epoch AI. Share solved.
- OpenRouter, fetched
- Epoch AI benchmarking hub, fetched