Math
FrontierMath Erdős Leaderboard
FrontierMath Erdős, from Epoch AI, tests 68 mathematical conjectures that were open when the benchmark was assembled in August 2026. Models must produce complete Lean proofs or disproofs that pass an isolated proof checker; the score is the fraction of conjectures resolved within the evaluation budget.
Source: epoch1 open models ranked+14 proprietaryData through Nov 2026
All models ranked on FrontierMath Erdős
Proprietary / closed models are shown dimmed — you can't run them locally, but they show where the open field stands.
| # | Model | Score | VRAM @ Q4 |
|---|---|---|---|
| 1 | GPT 6.1 Sol Max · proprietary | 4.4% | API only |
| 2 | Claude Opus 5.5 Max · proprietary | 2.9% | API only |
| 3 | Gemini 4 Carbon (high) · proprietary | 2.9% | API only |
| 4 | GPT 6 Astra Max · proprietary | 2.9% | API only |
| 5 | Claude Sonnet 5.5 Max · proprietary | 2.9% | API only |
| 6 | Gemini 4 Argon (high) · proprietary | 1.5% | API only |
| 7 | GPT 6 Sol Max · proprietary | 1.5% | API only |
| 8 | Claude Fable 5 Max · proprietary | 0.0% | API only |
| 9 | Claude Fable 5.1 Max · proprietary | 0.0% | API only |
| 10 | Claude Opus 5 Max · proprietary | 0.0% | API only |
| 11 | Gemini 3.8 Flash (high) · proprietary | 0.0% | API only |
| 12 | GPT 5.5 (xhigh) · proprietary | 0.0% | API only |
| 13 | GPT 5.6 Sol Max · proprietary | 0.0% | API only |
| 14 | Grok 4.7 (xhigh) · proprietary | 0.0% | API only |
| 15 | Kimi K3 · 2779.9B | 0.0% | ~1674 GB |
FrontierMath Erdős: frequently asked questions
- What is the best open LLM on FrontierMath Erdős?
- Kimi K3 is the top open model on FrontierMath Erdős, scoring 0.0%. Among all models tested — including proprietary ones — it ranks #8. The top model overall is GPT 6.1 Sol Max (OpenAI) at 4.4%.
- Can open models match proprietary models on FrontierMath Erdős?
- Not quite on FrontierMath Erdős: the strongest proprietary model (GPT 6.1 Sol Max) scores 4.4%, ahead of the best open model (Kimi K3) at 0.0% — but you can run the open one yourself.
Scores aggregated from epoch. llmrun does not run this benchmark — see the source for methodology, or the about benchmarks for what it measures.