Benchmark · Reasoning and knowledge
FrontierMath (Tiers 1-3)
Original research-level mathematics problems written and checked by mathematicians, answered by a Python function the model submits after reasoning with a Python tool. Epoch AI made the set and keeps most of it private; version 2 of June 2026 fixed errors in 42% of problems.
- Score
- % of problems solved, higher is better
- Run by
- Epoch AI
- Status
- Current, newest result 19 November 2026
- Updated
- as models are released
- Ceiling
- The best result is within five points of a perfect score, so the test no longer tells the top models apart
- Data
- the published file, CC BY 4.0, read 20 November 2026
Results: Tiers 1-3 v2 (2026-06-12)
36 models we list, each at its best setting, best first.
| # | Model | Score | Range | Who ran it | Setting | Date | |
|---|---|---|---|---|---|---|---|
| 1 | Claude Fable 5.2 A | 95.1% | Independent | effort max | 2026-11-06 | source | |
| 2 | GPT-6.1 Sol AA | 93.7% | Independent | effort max | 2026-09-29 | source | |
| 2 | GPT-6 Astra A | 93.7% | Independent | effort max | 2026-08-30 | source | |
| 4 | GPT-6.2 Astra AA | 92.6% | Independent | effort max | 2026-10-27 | source | |
| 5 | Gemini 4 Neon B | 91.2% | Independent | effort xhigh | 2026-11-19 | source | |
| 5 | Claude Opus 5.5 A | 91.2% | Independent | effort max | 2026-09-22 | source | |
| 7 | GPT-6.1 Terra AA | 90.9% | Independent | effort max | 2026-10-26 | source | |
| 8 | Gemini 4 Carbon B | 90.5% | Independent | effort high | 2026-11-02 | source | |
| 9 | Claude Fable 5.1 A | 90.2% | Independent | effort max | 2026-09-01 | source | |
| 10 | GPT-6 Sol A | 89.8% | Independent | effort max | 2026-09-22 | source | |
| 11 | Claude Sonnet 5.5 A | 88.8% | Independent | effort max | 2026-09-29 | source | |
| 12 | Gemini 4 Hydrogen B | 88.1% | Independent | effort xhigh | 2026-11-13 | source | |
| 13 | Gemini 4 Argon B | 86.0% | Independent | effort high | 2026-10-13 | source | |
| 14 | Muse Wonder D | 82.1% | Independent | effort max | 2026-10-12 | source | |
| 15 | GPT-6 Luna A | 79.0% | Independent | effort max | 2026-09-22 | source | |
| 16 | Kimi K3.1 Pro D | 78.2% | Independent | effort max | 2026-11-06 | source | |
| 17 | Claude Haiku 5.5 A | 75.1% | Independent | effort max | 2026-10-07 | source | |
| 18 | Muse Spark 1.3 D | 74.4% | Independent | effort xhigh | 2026-09-16 | source | |
| 19 | Kimi K3 B | 72.2% | Independent | effort max | 2026-07-17 | source | |
| 20 | DeepSeek V4.1 Flash C | 71.2% | Independent | effort max | 2026-10-19 | source | |
| 21 | Muse Spark Open C | 69.8% | Independent | effort xhigh | 2026-10-15 | source | |
| 22 | GLM-5.3 B | 68.8% | Independent | effort max | 2026-08-25 | source | |
| 23 | Gemini 3.8 Flash AA | 68.4% | Independent | effort high | 2026-09-02 | source | |
| 24 | Gemini 3.1 Pro (preview) AA | 59.6% | Independent | 2026-06-11 | source | ||
| 25 | DeepSeek-V4-Flash-0731 C | 57.5% | Independent | effort max | 2026-08-02 | source | |
| 26 | GLM-5.3-Flash B | 55.8% | Independent | effort max | 2026-08-27 | source | |
| 27 | Grok 4.7 BB | 53.0% | Independent | effort xhigh | 2026-09-22 | source | |
| 28 | Inkling-Small B | 46.3% | Independent | effort xhigh | 2026-08-14 | source | |
| 29 | DeepSeek V4 Pro D | 45.3% | Independent | effort max | 2026-06-17 | source | |
| 30 | Grok 4.3 BB | 42.8% | Independent | effort high | 2026-06-17 | source | |
| 31 | Qwen3.6-27B B | 35.1% | Independent | 2026-08-28 | source | ||
| 32 | Inkling B | 33.3% | Independent | effort xhigh | 2026-08-06 | source | |
| 33 | Qwen3.5-397B-A17B B | 31.2% | Independent | effort none | 2026-08-30 | source | |
| 34 | Gemini 3.1 Flash-Lite B | 27.7% | Independent | effort high | 2026-08-30 | source | |
| 35 | Gemini 3.5 Flash-Lite AA | 26.0% | Independent | effort high | 2026-08-02 | source | |
| 36 | Qwen3.6-35B-A3B B | 20.4% | Independent | effort none | 2026-08-30 | source |
Every setting, 43 rows
| Model | As the leaderboard names it | Score | Who ran it | Setting | Date | |
|---|---|---|---|---|---|---|
| Claude Fable 5.2 | claude-fable-5-2 | 95.1% | Independent | effort max | 2026-11-06 | source |
| GPT-6.1 Sol | gpt-6.1-sol | 93.7% | Independent | effort max | 2026-09-29 | source |
| GPT-6 Astra | gpt-6-astra | 93.7% | Independent | effort max | 2026-08-30 | source |
| GPT-6.2 Astra | gpt-6.2-astra | 92.6% | Independent | effort max | 2026-10-27 | source |
| Gemini 4 Neon | gemini-4-neon | 91.2% | Independent | effort xhigh | 2026-11-19 | source |
| Claude Opus 5.5 | claude-opus-5-5 | 91.2% | Independent | effort max | 2026-09-22 | source |
| GPT-6.1 Terra | gpt-6.1-terra | 90.9% | Independent | effort max | 2026-10-26 | source |
| Gemini 4 Carbon | gemini-4-carbon | 90.5% | Independent | effort high | 2026-11-02 | source |
| Claude Fable 5.1 | claude-fable-5-1 | 90.2% | Independent | effort max | 2026-09-01 | source |
| GPT-6 Sol | gpt-6-sol | 89.8% | Independent | effort max | 2026-09-22 | source |
| Claude Sonnet 5.5 | claude-sonnet-5-5 | 88.8% | Independent | effort max | 2026-09-29 | source |
| Gemini 4 Hydrogen | gemini-4-hydrogen | 88.1% | Independent | effort xhigh | 2026-11-13 | source |
| Gemini 4 Argon | gemini-4-argon | 86.0% | Independent | effort high | 2026-10-13 | source |
| Muse Wonder | muse-wonder-1.0 | 82.1% | Independent | effort max | 2026-10-12 | source |
| GPT-6 Luna | gpt-6-luna | 79.0% | Independent | effort max | 2026-09-22 | source |
| Kimi K3.1 Pro | kimi-k3.1-pro-together | 78.2% | Independent | effort max | 2026-11-06 | source |
| Claude Haiku 5.5 | claude-haiku-5-5 | 75.1% | Independent | effort max | 2026-10-07 | source |
| Muse Spark 1.3 | muse-spark-1.3 | 74.4% | Independent | effort xhigh | 2026-09-16 | source |
| Muse Spark 1.3 | muse-spark-1.3 | 74.0% | Independent | effort max | 2026-09-18 | source |
| Kimi K3 | kimi-k3 | 72.2% | Independent | effort max | 2026-07-17 | source |
| DeepSeek V4.1 Flash | deepseek-v4.1-flash-1017 | 71.2% | Independent | effort max | 2026-10-19 | source |
| Muse Spark Open | Muse-Spark-Open-816B-A42B | 69.8% | Independent | effort xhigh | 2026-10-15 | source |
| GLM-5.3 | glm-5.3 | 68.8% | Independent | effort max | 2026-08-25 | source |
| Gemini 3.8 Flash | gemini-3.8-flash | 68.4% | Independent | effort high | 2026-09-02 | source |
| DeepSeek V4.1 Flash | deepseek-v4.1-flash | 67.4% | Independent | effort max | 2026-10-08 | source |
| Gemini 3.1 Pro (preview) | gemini-3.1-pro-preview | 59.6% | Independent | 2026-06-11 | source | |
| DeepSeek-V4-Flash-0731 | deepseek-v4-flash-0731 | 57.5% | Independent | effort max | 2026-08-02 | source |
| GLM-5.3-Flash | glm-5.3-flash | 55.8% | Independent | effort max | 2026-08-27 | source |
| Grok 4.7 | grok-4.7 | 53.0% | Independent | effort xhigh | 2026-09-22 | source |
| Inkling-Small | inkling-small | 46.3% | Independent | effort xhigh | 2026-08-14 | source |
| DeepSeek V4 Pro | deepseek-v4-pro | 45.3% | Independent | effort max | 2026-06-17 | source |
| Grok 4.3 | grok-4.3 | 42.8% | Independent | effort high | 2026-06-17 | source |
| Qwen3.6-27B | qwen3.6-27b | 35.1% | Independent | 2026-08-28 | source | |
| Qwen3.6-27B | qwen3.6-27b | 34.0% | Independent | effort none | 2026-08-29 | source |
| Inkling | Inkling | 33.3% | Independent | effort xhigh | 2026-08-06 | source |
| Qwen3.5-397B-A17B | qwen3.5-397b-a17b | 31.2% | Independent | effort none | 2026-08-30 | source |
| Qwen3.5-397B-A17B | qwen3.5-397b-a17b | 29.5% | Independent | 2026-08-28 | source | |
| Gemini 3.1 Flash-Lite | gemini-3.1-flash-lite | 27.7% | Independent | effort high | 2026-08-30 | source |
| Gemini 3.5 Flash-Lite | gemini-3.5-flash-lite | 26.0% | Independent | effort high | 2026-08-02 | source |
| Gemini 3.1 Flash-Lite | gemini-3.1-flash-lite | 22.5% | Independent | effort low | 2026-08-30 | source |
| Gemini 3.1 Flash-Lite | gemini-3.1-flash-lite | 21.4% | Independent | effort minimal | 2026-08-28 | source |
| Qwen3.6-35B-A3B | qwen3.6-35b-a3b | 20.4% | Independent | effort none | 2026-08-30 | source |
| Qwen3.6-35B-A3B | qwen3.6-35b-a3b | 17.5% | Independent | 2026-08-29 | source |
Results: Tiers 1-3 (2025-02-28)
1 model we list, each at its best setting, best first.
| # | Model | Score | Range | Who ran it | Setting | Date | |
|---|---|---|---|---|---|---|---|
| 1 | Gemini 3.1 Pro (preview) AA | 36.9% | Independent | 2026-02-19 | source |
A rank is among the models we list with a result, not a place on the whole leaderboard, and a model offered by several hosts is shown once, on its own listing. These results don't move our grades. Data: Epoch AI, Capabilities and benchmarking (CC BY 4.0), CC BY 4.0.
For companies
Do agents find, use and choose your tools?
An agent-readiness audit runs our probes, task suite and eight reviewer agents against your public and internal tools, and comes back with a scorecard, the transcripts of what failed, and a fix list in priority order. From $2,500, re-run included. We never take payment to move a rank. We do help companies earn one.