FrontierMath Tiers 1–3 (v2)
Expert-written math problems from advanced undergraduate level to early career research.
- Unit
- % solved
- Results
- 120 results86 Tokscale models matched, 87 in the source
- Last update
- Oct 15, 2026Newest run date
- Retrieved
- Oct 21, 2026Source data through Oct 15, 2026
Conditions. Run by Epoch AI at each reasoning effort.
Attribution. Epoch AI, ‘Capabilities & benchmarking’, epoch.ai/benchmarks (CC BY 4.0)
Near its ceiling: the top score is 93.7%, and Epoch AI estimates the reachable maximum at 95%. Differences between top models are small.
Results
One row per model: its best reported result. The rank is the position in this list, by score; the source's own ranking can differ.
Tokscale usage counts accounts that used the model in public Tokscale submissions, Oct 14 – Nov 12, 2026. It is shown next to the score and does not explain it.
| Rank | Model | Score (% solved) | Conditions | Date | Tokscale usage (30 days) | Source link |
|---|---|---|---|---|---|---|
| 1 | GPT-6.1 SolListed as gpt-6.1-sol_max | 93.7%Standard error ±1.4 pts | Max effort | Sep 29, 2026Run date | 661active users | Source for GPT-6.1 Sol (opens in a new tab) |
| 1 | GPT-6 AstraListed as gpt-6-astra_max | 93.7%Standard error ±1.4 pts | Max effort | Aug 30, 2026Run date | 563active users | Source for GPT-6 Astra (opens in a new tab) |
| 3 | Claude Opus 5.5Listed as claude-opus-5-5_max | 91.2%Standard error ±1.7 pts | Max effort | Sep 22, 2026Run date | 711active users | Source for Claude Opus 5.5 (opens in a new tab) |
| 4 | Claude Fable 5.1Listed as claude-fable-5-1_max | 90.2%Standard error ±1.8 pts | Max effort | Sep 1, 2026Run date | 361active users | Source for Claude Fable 5.1 (opens in a new tab) |
| 5 | GPT-6 SolListed as gpt-6-sol_max | 89.8%Standard error ±1.8 pts | Max effort | Sep 22, 2026Run date | 246active users | Source for GPT-6 Sol (opens in a new tab) |
| 6 | GPT-5.6 SolListed as gpt-5.6-sol_max | 89.1%Standard error ±1.8 pts | Max effort | Jul 9, 2026Run date | 218active users | Source for GPT-5.6 Sol (opens in a new tab) |
| 7 | Claude Sonnet 5.5Listed as claude-sonnet-5-5_max | 88.8%Standard error ±1.9 pts | Max effort | Sep 29, 2026Run date | 555active users | Source for Claude Sonnet 5.5 (opens in a new tab) |
| 8 | GPT-5.5 ProListed as gpt-5.5-pro_xhigh | 87.7%Standard error ±1.9 pts | XHigh effort | Jun 12, 2026Run date | 1active users | Source for GPT-5.5 Pro (opens in a new tab) |
| 9 | Claude Fable 5Listed as claude-fable-5_max | 87.0%Standard error ±2.0 pts | Max effort | Jun 9, 2026Run date | 41active users | Source for Claude Fable 5 (opens in a new tab) |
| 10 | Gemini 4 ArgonListed as gemini-4-argon_high | 86.0%Standard error ±2.1 pts | High effort | Oct 13, 2026Run date | 82active users | Source for Gemini 4 Argon (opens in a new tab) |
| 10 | GPT-5.6 TerraListed as gpt-5.6-terra_max | 86.0%Standard error ±2.1 pts | Max effort | Jul 9, 2026Run date | 121active users | Source for GPT-5.6 Terra (opens in a new tab) |
| 12 | Claude Opus 5Listed as claude-opus-5_max | 85.6%Standard error ±2.1 pts | Max effort | Jul 24, 2026Run date | 163active users | Source for Claude Opus 5 (opens in a new tab) |
| 13 | GPT-5.5Listed as gpt-5.5_xhigh | 85.3%Standard error ±2.1 pts | XHigh effort | Jun 11, 2026Run date | 30active users | Source for GPT-5.5 (opens in a new tab) |
| 14 | GPT-5.4 ProListed as gpt-5.4-pro-2026-03-05_xhigh | 82.5%Standard error ±2.3 pts | XHigh effort | Jun 13, 2026Run date | No Tokscale usage in the last 30 days | Source for GPT-5.4 Pro (opens in a new tab) |
| 15 | Muse Wonder 1.0Listed as muse-wonder-1.0_max | 82.1%Standard error ±2.3 pts | Max effort | Oct 12, 2026Run date | 17active users | Source for Muse Wonder 1.0 (opens in a new tab) |
| 15 | GPT-5.6 LunaListed as gpt-5.6-luna_max | 82.1%Standard error ±2.3 pts | Max effort | Jul 9, 2026Run date | 159active users | Source for GPT-5.6 Luna (opens in a new tab) |
| 17 | Claude Opus 4.8Listed as claude-opus-4-8_max | 80.0%Standard error ±2.4 pts | Max effort | Jun 10, 2026Run date | 222active users | Source for Claude Opus 4.8 (opens in a new tab) |
| 18 | GPT-6 LunaListed as gpt-6-luna_max | 78.9%Standard error ±2.4 pts | Max effort | Sep 22, 2026Run date | 433active users | Source for GPT-6 Luna (opens in a new tab) |
| 19 | GPT-5.4Listed as gpt-5.4-2026-03-05_xhigh | 78.6%Standard error ±2.4 pts | XHigh effort | Jun 11, 2026Run date | 1active users | Source for GPT-5.4 (opens in a new tab) |
| 20 | Claude Haiku 5.5Listed as claude-haiku-5-5_max | 75.1%Standard error ±2.6 pts | Max effort | Oct 7, 2026Run date | 477active users | Source for Claude Haiku 5.5 (opens in a new tab) |
| 21 | Qwen3.8 MaxListed as qwen3.8-max_xhigh | 74.7%Standard error ±2.6 pts | XHigh effort | Aug 4, 2026Run date | 16active users | Source for Qwen3.8 Max (opens in a new tab) |
| 22 | Muse Spark 1.3Listed as muse-spark-1.3_xhigh | 74.4%Standard error ±2.6 pts | XHigh effort | Sep 16, 2026Run date | 21active users | Source for Muse Spark 1.3 (opens in a new tab) |
| 23 | GPT-5.2 ProListed as gpt-5.2-pro-2025-12-11_xhigh | 74.0%Standard error ±2.6 pts | XHigh effort | Jun 13, 2026Run date | No Tokscale usage in the last 30 days | Source for GPT-5.2 Pro (opens in a new tab) |
| 24 | Kimi K3Listed as kimi-k3_max | 72.2%Standard error ±2.7 pts | Max effort | Jul 17, 2026Run date | 56active users | Source for Kimi K3 (opens in a new tab) |
| 25 | Gemini 3.7 FlashListed as gemini-3.7-flash_high | 71.6%Standard error ±2.7 pts | High effort | Aug 14, 2026Run date | 4active users | Source for Gemini 3.7 Flash (opens in a new tab) |
| 26 | Claude Opus 4.7Listed as claude-opus-4-7_max | 70.2%Standard error ±2.7 pts | Max effort | Jun 10, 2026Run date | 27active users | Source for Claude Opus 4.7 (opens in a new tab) |
| 27 | Muse Spark OpenListed as Muse-Spark-Open-816B-A42B_xhigh | 69.8%Standard error ±2.7 pts | XHigh effort | Oct 15, 2026Run date | 2active users | Source for Muse Spark Open (opens in a new tab) |
| 28 | GLM-5.3Listed as glm-5.3_max | 68.8%Standard error ±2.7 pts | Max effort | Aug 25, 2026Run date | 139active users | Source for GLM-5.3 (opens in a new tab) |
| 29 | Gemini 3.8 FlashListed as gemini-3.8-flash_high | 68.4%Standard error ±2.8 pts | High effort | Sep 2, 2026Run date | 66active users | Source for Gemini 3.8 Flash (opens in a new tab) |
| 30 | GPT-5.2Listed as gpt-5.2-2025-12-11_xhigh | 67.4%Standard error ±2.8 pts | XHigh effort | Jun 11, 2026Run date | No Tokscale usage in the last 30 days | Source for GPT-5.2 (opens in a new tab) |
| 31 | DeepSeek V4.1 FlashListed as deepseek-v4.1-flash_max | 67.4%Standard error ±2.8 pts | Max effort | Oct 8, 2026Run date | 225active users | Source for DeepSeek V4.1 Flash (opens in a new tab) |
| 32 | Grok 4.6Listed as grok-4.6_xhigh | 66.0%Standard error ±2.8 pts | XHigh effort | Aug 14, 2026Run date | 41active users | Source for Grok 4.6 (opens in a new tab) |
| 32 | Claude Opus 4.6Listed as claude-opus-4-6_max | 66.0%Standard error ±2.8 pts | Max effort | Jun 11, 2026Run date | 52active users | Source for Claude Opus 4.6 (opens in a new tab) |
| 34 | Qwen3.8 Max 0902Listed as qwen3.8-max-0902_xhigh | 65.6%Standard error ±2.8 pts | XHigh effort | Sep 2, 2026Run date | 1active users | Source for Qwen3.8 Max 0902 (opens in a new tab) |
| 34 | Claude Sonnet 5Listed as claude-sonnet-5_max | 65.6%Standard error ±2.8 pts | Max effort | Jun 30, 2026Run date | 138active users | Source for Claude Sonnet 5 (opens in a new tab) |
| 36 | DeepSeek V4 Pro 0813Listed as deepseek-v4-pro-0813_max | 64.6%Standard error ±2.8 pts | Max effort | Aug 19, 2026Run date | 6active users | Source for DeepSeek V4 Pro 0813 (opens in a new tab) |
| 36 | Qwen3.7 MaxListed as qwen3.7-max | 64.6%Standard error ±2.8 pts | Not reported | Jun 13, 2026Run date | 3active users | Source for Qwen3.7 Max (opens in a new tab) |
| 38 | Gemini 3.5 FlashListed as gemini-3.5-flash_high | 62.8%Standard error ±2.9 pts | High effort | Jun 10, 2026Run date | 3active users | Source for Gemini 3.5 Flash (opens in a new tab) |
| 39 | Gemini 3.1 Pro PreviewListed as gemini-3.1-pro-preview | 59.6%Standard error ±2.9 pts | Not reported | Jun 11, 2026Run date | 5active users | Source for Gemini 3.1 Pro Preview (opens in a new tab) |
| 40 | GLM-5.2Listed as glm-5.2_max | 59.2%Standard error ±3.0 pts | Max effort | Jun 19, 2026Run date | 13active users | Source for GLM-5.2 (opens in a new tab) |
| 41 | Gemini 3.6 FlashListed as gemini-3.6-flash_high | 58.9%Standard error ±2.9 pts | High effort | Aug 2, 2026Run date | 3active users | Source for Gemini 3.6 Flash (opens in a new tab) |
| 42 | Mistral Large 4Listed as mistral-large-4-0_high | 58.6%Standard error ±2.9 pts | High effort | Oct 12, 2026Run date | 4active users | Source for Mistral Large 4 (opens in a new tab) |
| 43 | DeepSeek V4 Flash 0731Listed as deepseek-v4-flash-0731_max | 57.5%Standard error ±2.9 pts | Max effort | Aug 2, 2026Run date | 18active users | Source for DeepSeek V4 Flash 0731 (opens in a new tab) |
| 44 | Grok 4.5Listed as grok-4.5_high | 57.2%Standard error ±2.9 pts | High effort | Jul 9, 2026Run date | 9active users | Source for Grok 4.5 (opens in a new tab) |
| 44 | Kimi K2.6Listed as kimi-k2.6 | 57.2%Standard error ±2.9 pts | Not reported | Jun 10, 2026Run date | 4active users | Source for Kimi K2.6 (opens in a new tab) |
| 46 | GLM-5.3 FlashListed as glm-5.3-flash_max | 55.8%Standard error ±2.9 pts | Max effort | Aug 27, 2026Run date | 170active users | Source for GLM-5.3 Flash (opens in a new tab) |
| 46 | GPT-5 ProListed as gpt-5-pro-2025-10-06_high | 55.8%Standard error ±2.9 pts | High effort | Jun 12, 2026Run date | No Tokscale usage in the last 30 days | Source for GPT-5 Pro (opens in a new tab) |
| 48 | GPT-5 2025 08 07Listed as gpt-5-2025-08-07_high | 55.4%Standard error ±2.9 pts | High effort | Jun 10, 2026Run date | No Tokscale usage in the last 30 days | Source for GPT-5 2025 08 07 (opens in a new tab) |
| 49 | Kimi K2.7 CodeListed as kimi-k2.7-code | 54.0%Standard error ±3.0 pts | Not reported | Jun 13, 2026Run date | 17active users | Source for Kimi K2.7 Code (opens in a new tab) |
| 50 | Grok 4.7Listed as grok-4.7_xhigh | 53.0%Standard error ±3.0 pts | XHigh effort | Sep 22, 2026Run date | 131active users | Source for Grok 4.7 (opens in a new tab) |
Showing 50 of 87 models
Score by release date
Each dot is one model's best result on FrontierMath Tiers 1–3 (v2), placed at the model's release date. The dashed line is the best score among models released up to that date.
- Used on Tokscale in the last 30 days
- No Tokscale usage in the last 30 days
- Best score so far
GPT-6 Astra: 93.7% · released Sep 3, 2026 · Max effort · 563 active users on Tokscale in the last 30 days
86 models, released Apr 9, 2024 to Oct 15, 2026. Highest: GPT-6 Astra, 93.7%, released Sep 3, 2026. Use the left and right arrow keys to read each model in release order.0%20%40%60%80%100%
Show data
| Model | Released | Score | Tokscale users, 30 days |
|---|---|---|---|
| Muse Spark Open | Oct 15, 2026 | 69.8%XHigh effort | 2 |
| Muse Wonder 1.0 | Oct 12, 2026 | 82.1%Max effort | 17 |
| Claude Haiku 5.5 | Oct 7, 2026 | 75.1%Max effort | 477 |
| Mistral Large 4 | Oct 6, 2026 | 58.6%High effort | 4 |
| Gemini 4 Argon | Sep 30, 2026 | 86.0%High effort | 82 |
| GPT-6.1 Sol | Sep 29, 2026 | 93.7%Max effort | 661 |
| Claude Sonnet 5.5 | Sep 28, 2026 | 88.8%Max effort | 555 |
| GPT-6 Luna | Sep 22, 2026 | 78.9%Max effort | 433 |
| GPT-6 Sol | Sep 22, 2026 | 89.8%Max effort | 246 |
| Claude Opus 5.5 | Sep 22, 2026 | 91.2%Max effort | 711 |
| Grok 4.7 | Sep 21, 2026 | 53.0%XHigh effort | 131 |
| DeepSeek V4.1 Flash | Sep 9, 2026 | 67.4%Max effort | 225 |
| GPT-6 Astra | Sep 3, 2026 | 93.7%Max effort | 563 |
| Gemini 3.8 Flash | Sep 2, 2026 | 68.4%High effort | 66 |
| Muse Spark 1.3 | Sep 2, 2026 | 74.4%XHigh effort | 21 |
| Qwen3.8 Max 0902 | Sep 1, 2026 | 65.6%XHigh effort | 1 |
| Claude Fable 5.1 | Sep 1, 2026 | 90.2%Max effort | 361 |
| GLM-5.3 Flash | Aug 20, 2026 | 55.8%Max effort | 170 |
| GLM-5.3 | Aug 14, 2026 | 68.8%Max effort | 139 |
| DeepSeek V4 Pro 0813 | Aug 13, 2026 | 64.6%Max effort | 6 |
| Gemini 3.7 Flash | Aug 13, 2026 | 71.6%High effort | 4 |
| Grok 4.6 | Aug 12, 2026 | 66.0%XHigh effort | 41 |
| Qwen3.8 Max | Aug 2, 2026 | 74.7%XHigh effort | 16 |
| DeepSeek V4 Flash 0731 | Jul 31, 2026 | 57.5%Max effort | 18 |
| Qwen3.7 Flash | Jul 27, 2026 | 19.3% | No Tokscale usage in the last 30 days |
| Claude Opus 5 | Jul 24, 2026 | 85.6%Max effort | 163 |
| Gemini 3.5 Flash Lite | Jul 21, 2026 | 26.0%High effort | 13 |
| Gemini 3.6 Flash | Jul 21, 2026 | 58.9%High effort | 3 |
| Kimi K3 | Jul 16, 2026 | 72.2%Max effort | 56 |
| Inkling | Jul 15, 2026 | 33.3%XHigh effort | No Tokscale usage in the last 30 days |
| Inkling Small | Jul 15, 2026 | 46.3%XHigh effort | No Tokscale usage in the last 30 days |
| GPT-5.6 Luna | Jul 9, 2026 | 82.1%Max effort | 159 |
| GPT-5.6 Terra | Jul 9, 2026 | 86.0%Max effort | 121 |
| GPT-5.6 Sol | Jul 9, 2026 | 89.1%Max effort | 218 |
| Grok 4.5 | Jul 8, 2026 | 57.2%High effort | 9 |
| Claude Sonnet 5 | Jun 30, 2026 | 65.6%Max effort | 138 |
| GLM-5.2 | Jun 16, 2026 | 59.2%Max effort | 13 |
| Kimi K2.7 Code | Jun 12, 2026 | 54.0% | 17 |
| Claude Fable 5 | Jun 9, 2026 | 87.0%Max effort | 41 |
| Qwen3.7 Plus | Jun 2, 2026 | 34.4%No reasoning | 3 |
| Claude Opus 4.8 | May 28, 2026 | 80.0%Max effort | 222 |
| Gemini 3.5 Flash | May 19, 2026 | 62.8%High effort | 3 |
| Qwen3.7 Max | May 19, 2026 | 64.6% | 3 |
| GPT-5.5 Instant | May 5, 2026 | 26.3% | No Tokscale usage in the last 30 days |
| Qwen3.6 Flash | Apr 27, 2026 | 22.5%No reasoning | No Tokscale usage in the last 30 days |
| DeepSeek V4 Pro | Apr 24, 2026 | 45.3%Max effort | 51 |
| GPT-5.5 | Apr 23, 2026 | 85.3%XHigh effort | 30 |
| GPT-5.5 Pro | Apr 23, 2026 | 87.7%XHigh effort | 1 |
| Qwen3.6 27B | Apr 22, 2026 | 35.1% | No Tokscale usage in the last 30 days |
| Kimi K2.6 | Apr 20, 2026 | 57.2% | 4 |
| Grok 4.3 | Apr 17, 2026 | 42.8%High effort | No Tokscale usage in the last 30 days |
| Claude Opus 4.7 | Apr 16, 2026 | 70.2%Max effort | 27 |
| Qwen3.6 35B A3B | Apr 14, 2026 | 20.4%No reasoning | 1 |
| GLM-5.1 | Apr 7, 2026 | 36.8% | 5 |
| Qwen3.6 Plus | Mar 31, 2026 | 38.2% | 1 |
| GPT-5.4 Nano | Mar 17, 2026 | 44.9%High effort | 4 |
| GPT-5.4 Mini | Mar 17, 2026 | 51.2%XHigh effort | 1 |
| GPT-5.4 | Mar 5, 2026 | 78.6%XHigh effort | 1 |
| GPT-5.4 Pro | Mar 5, 2026 | 82.5%XHigh effort | No Tokscale usage in the last 30 days |
| Gemini 3.1 Flash Lite | Mar 3, 2026 | 27.7%High effort | 5 |
| Qwen3.5 Flash | Feb 25, 2026 | 18.2%No reasoning | No Tokscale usage in the last 30 days |
| Gemini 3.1 Pro Preview | Feb 19, 2026 | 59.6% | 5 |
| Grok 4.20 0309 Reasoning | Feb 17, 2026 | 44.9% | 1 |
| Qwen3.5 397B A17B | Feb 13, 2026 | 31.2%No reasoning | No Tokscale usage in the last 30 days |
| Claude Opus 4.6 | Feb 5, 2026 | 66.0%Max effort | 52 |
| Gemini 3 Flash Preview | Dec 17, 2025 | 51.2% | 4 |
| GPT-5.2 | Dec 11, 2025 | 67.4%XHigh effort | No Tokscale usage in the last 30 days |
| GPT-5.2 Pro | Dec 11, 2025 | 74.0%XHigh effort | No Tokscale usage in the last 30 days |
| Claude Opus 4.5 | Nov 24, 2025 | 34.4%32K thinking budget | 4 |
| GPT-5 Pro | Oct 7, 2025 | 55.8%High effort | No Tokscale usage in the last 30 days |
| Claude Sonnet 4.5 | Sep 29, 2025 | 23.9%32K thinking budget | 23 |
| Qwen3 Max | Sep 24, 2025 | 18.9% | No Tokscale usage in the last 30 days |
| GPT-5 Nano | Aug 7, 2025 | 20.0%High effort | No Tokscale usage in the last 30 days |
| GPT-5 Mini | Aug 7, 2025 | 46.7%High effort | 1 |
| GPT-5 2025 08 07 | Aug 7, 2025 | 55.4%High effort | No Tokscale usage in the last 30 days |
| Claude Opus 4.1 | Aug 5, 2025 | 12.6%32K thinking budget | No Tokscale usage in the last 30 days |
| Gemini 2.5 Pro | Jun 17, 2025 | 24.6% | 1 |
| o3 | Apr 16, 2025 | 33.3%High effort | No Tokscale usage in the last 30 days |
| o4-mini | Apr 16, 2025 | 36.1%High effort | No Tokscale usage in the last 30 days |
| GPT-4.1 2025 04 14 | Apr 14, 2025 | 6.0% | 5 |
| GPT-4.1 Mini 2025 04 14 | Apr 14, 2025 | 6.7% | 5 |
| o3-mini | Jan 31, 2025 | 18.6%High effort | No Tokscale usage in the last 30 days |
| o1 | Dec 17, 2024 | 14.7%High effort | No Tokscale usage in the last 30 days |
| GPT-4o 2024 08 06 | Aug 6, 2024 | 0.4% | No Tokscale usage in the last 30 days |
| GPT-4o Mini | Jul 18, 2024 | 0.7% | 1 |
| GPT-4 Turbo | Apr 9, 2024 | 0.7% | No Tokscale usage in the last 30 days |
Method and matching
How results are run
- Run by Epoch AI at each reasoning effort.
Matching to Tokscale models
- 119 of 120 results are matched to 86 Tokscale models: 82 by exact id, 37 by normalized name.
- No match is marked as reviewed by hand yet.
- 1 result from 1 model is not matched. They stay in the table without a Tokscale link.
- Reasoning effort, thinking budget and agent are conditions of a result. They never change which model a result belongs to.
License and attribution
- CC BY 4.0 (opens in a new tab), retrieved Oct 21, 2026.
- Epoch AI, ‘Capabilities & benchmarking’, epoch.ai/benchmarks (CC BY 4.0)
Sources and attribution
Each score keeps its own source, unit and run conditions. Tokscale does not combine benchmarks into one score, and a missing score means the model was not evaluated, not that it scored zero.