FrontierMath-Erdos
Re-run by Epoch AINo description yet: we only describe an eval set when we can do it correctly.
- Kind
- Run independently by Epoch AI, with public logs. Our own re-runs are not built yet.
- Score shown
- Epoch's column "Best score (across scorers)", as a percentage, higher is better. Error bars as the source gives them.
- Results
- 16 in all: 1 for open-weight models, 15 for closed ones. 1 linked to a model page here.
Best open-weight result: moonshotai/Kimi-K3, 0.0%, number 16 of 16 results.
| # | Model | Weights | Setting | Score | Date | Source |
|---|---|---|---|---|---|---|
| 1 | gpt-6.2-astra_maxOpenAI | closed | effort max | 10.3%±3.7 | 28 Oct 2026 | Epoch AI |
| 2 | gpt-6.1-sol_maxOpenAI | closed | effort max | 4.4%±2.5 | 1 Oct 2026 | Epoch AI |
| 3 | gpt-6-astra_maxOpenAI | closed | effort max | 2.9%±2.1 | 28 Aug 2026 | Epoch AI |
| 4 | claude-opus-5-5_maxAnthropic | closed | effort max | 2.9%±2.1 | 24 Sep 2026 | Epoch AI |
| 5 | gemini-4-carbon_highGoogle DeepMind | closed | effort high | 2.9%±2.1 | 7 Nov 2026 | Epoch AI |
| 6 | claude-sonnet-5-5_maxAnthropic | closed | effort max | 2.9%±2.1 | 8 Oct 2026 | Epoch AI |
| 7 | gemini-4-argon_highGoogle DeepMind | closed | effort high | 1.5%±1.5 | 20 Oct 2026 | Epoch AI |
| 8 | gpt-6-sol_maxOpenAI | closed | effort max | 1.5%±1.5 | 30 Sep 2026 | Epoch AI |
| 9 | claude-fable-5_maxAnthropic | closed | effort max | 0.0% | 28 Aug 2026 | Epoch AI |
| 10 | claude-fable-5-1_maxAnthropic | closed | effort max | 0.0% | 1 Sep 2026 | Epoch AI |
| 11 | claude-opus-5_maxAnthropic | closed | effort max | 0.0% | 28 Aug 2026 | Epoch AI |
| 12 | gemini-3.8-flash_highGoogle DeepMind | closed | effort high | 0.0% | 2 Sep 2026 | Epoch AI |
| 13 | gpt-5.5_xhighOpenAI | closed | effort xhigh | 0.0% | 28 Aug 2026 | Epoch AI |
| 14 | gpt-5.6-sol_maxOpenAI | closed | effort max | 0.0% | 28 Aug 2026 | Epoch AI |
| 15 | grok-4.7_xhighxAI | closed | effort xhigh | 0.0% | 21 Sep 2026 | Epoch AI |
| 16 | moonshotai/Kimi-K3Epoch: kimi-k3_max, Moonshot | open | effort max | 0.0% | 28 Aug 2026 | Epoch AI |
Open-weight rows are shaded. Dates are when the result was run or evaluated, where the data says. Data from Epoch AI (CC-BY 4.0), fetched 19 Nov 2026: Epoch AI, ‘Capabilities & benchmarking’. Published online at epoch.ai. Retrieved from ‘https://epoch.ai/benchmarks’ [online resource]. Epoch AI. Method