FrontierMath v2 · Tier 4
Epoch AI / Reasoning · More difficult mathematical research problems
Across e/accCounts toward ranking
Evidence budget1.2%
Upstream data time10/16 02:52
Last successful sync10/17 08:05
What it measures and how
Store Tier 4 separately, sharing the mathematical evidence budget with Tiers 1–3. Retain sampling uncertainty.
How this evidence is used
Counts toward ranking; benchmarks in the same task family share a fixed budget.
Benchmark results
Fixed rules select one representative configuration per public model. Anonymous test models are omitted. Original source ranks are retained, with up to 30 models per benchmark.
| Source rank | Source model name | Raw result | Representative configuration |
|---|---|---|---|
| 1 | gpt-6.1-sol_maxOpenAI | 100.0% | Max reasoning |
| 2 | gpt-6-astra_maxOpenAI | 97.6% | Max reasoning |
| 6 | claude-opus-5-5_maxAnthropic | 95.0% | Max reasoning |
| 7 | claude-fable-5_maxAnthropic | 90.2% | Max reasoning |
| 8 | gpt-6-sol_maxOpenAI | 90.0% | Max reasoning |
| 9 | claude-fable-5-1_maxAnthropic | 87.8% | Max reasoning |
| 11 | gpt-5.6-sol_maxOpenAI | 82.9% | Max reasoning |
| 13 | claude-sonnet-5-5_maxAnthropic | 80.5% | Max reasoning |
| 15 | gemini-4-argon_highGoogle | 78.0% | High reasoning |
| 15 | gpt-5.5-pro_xhighOpenAI | 78.0% | xHigh reasoning |
| 18 | claude-opus-5_maxAnthropic | 73.2% | Max reasoning |
| 19 | gpt-5.5_xhighOpenAI | 72.5% | xHigh reasoning |
| 20 | gpt-5.6-terra_maxOpenAI | 70.7% | Max reasoning |
| 20 | muse-wonder-1.0_maxMeta AI | 70.7% | Max reasoning |
| 22 | gpt-5.6-luna_maxOpenAI | 61.0% | Max reasoning |
| 23 | gpt-5.4-pro-2026-03-05_xhighOpenAI | 58.5% | xHigh reasoning |
| 24 | claude-opus-4-8_maxAnthropic | 56.1% | Max reasoning |
| 24 | gpt-6-luna_maxOpenAI | 56.1% | Max reasoning |
| 26 | gpt-5.4-2026-03-05_xhighOpenAI | 49.0% | xHigh reasoning |
| 27 | muse-spark-1.3_maxMeta | 46.3% | Max reasoning |
| 27 | qwen3.8-max_xhighAlibaba | 46.3% | xHigh reasoning |
| 27 | claude-haiku-5-5_maxAnthropic | 46.3% | Max reasoning |
| 30 | gpt-5.2-pro-2025-12-11_xhighOpenAI | 46.0% | xHigh reasoning |
| 32 | kimi-k3_maxMoonshot AI | 39.0% | Max reasoning |
| 33 | gemini-3.7-flash_highGoogle | 36.6% | High reasoning |
| 34 | qwen3.8-max-0902_xhighAlibaba | 34.1% | xHigh reasoning |
| 34 | qwen3.7-maxAlibaba | 34.1% | Source default configuration |
| 36 | Muse-Spark-Open-816B-A42B_xhighMeta AI | 31.7% | xHigh reasoning |
| 36 | claude-opus-4-7_maxAnthropic | 31.7% | Max reasoning |
| 36 | grok-4.6_xhighxAI | 31.7% | xHigh reasoning |
Limitations and attribution
Counts toward ranking; benchmarks in the same task family share a fixed budget.
Data license: CC BY 4.0 · Data evaluated by Epoch AI itself
Results are published by Epoch AI. Raw scores and the e/acc consensus index use different scales and cannot be added directly.