ProofBench
ReportedNo description yet: we only describe an eval set when we can do it correctly.
- Kind
- Published by a third party's leaderboard, collected by Epoch AI. The leaderboard is named and linked.
- Link
- https://www.vals.ai/benchmarks/proof_bench
- Score shown
- Epoch's column "Accuracy", as a percentage, higher is better; random guessing scores about 0.0%. Error bars as the source gives them.
- Results
- 85 in all: 30 for open-weight models, 55 for closed ones. 24 linked to a model page here.
Best open-weight result: moonshotai/Kimi-K3.1-Pro, 91.0%, number 10 of 85 results.
| # | Model | Weights | Setting | Score | Date | Source |
|---|---|---|---|---|---|---|
| 1 | claude-fable-5-1_maxAnthropic | closed | effort max | 100.0% | not dated | vals.ai |
| 2 | claude-fable-5-2_maxAnthropic | closed | effort max | 100.0% | not dated | vals.ai |
| 3 | claude-opus-5-5_maxAnthropic | closed | effort max | 100.0% | not dated | vals.ai |
| 4 | claude-sonnet-5-5_maxAnthropic | closed | effort max | 100.0% | not dated | vals.ai |
| 5 | claude-opus-5_maxAnthropic | closed | effort max | 99.0%±1.0 | not dated | vals.ai |
| 6 | gemini-4-argon_unknownGoogle DeepMind | closed | effort unknown | 99.0% | not dated | vals.ai |
| 7 | gpt-6-astra_unknownOpenAI | closed | effort unknown | 99.0% | not dated | vals.ai |
| 8 | gpt-6.1-sol_unknownOpenAI | closed | effort unknown | 99.0% | not dated | vals.ai |
| 9 | claude-fable-5_maxAnthropic | closed | effort max; Reasoning effort: max | 95.0%±2.2 | not dated | vals.ai |
| 10 | moonshotai/Kimi-K3.1-ProEpoch: kimi-k3.1-pro_unknown, Moonshot | open | effort unknown | 91.0%±2.9 | not dated | vals.ai |
| 11 | claude-haiku-5-5_maxAnthropic | closed | effort max | 89.0% | not dated | vals.ai |
| 12 | moonshotai/Kimi-K3Epoch: kimi-k3_unknown, Moonshot | open | effort unknown | 87.0%±3.4 | not dated | vals.ai |
| 13 | muse-wonder-1.0_unknownMeta AI | closed | effort unknown | 84.0% | not dated | vals.ai |
| 14 | gpt-5.6-sol_maxOpenAI | closed | effort max; Reasoning effort: max | 83.0%±3.8 | not dated | vals.ai |
| 15 | gpt-6-sol_unknownOpenAI | closed | effort unknown | 83.0% | not dated | vals.ai |
| 16 | claude-sonnet-5_maxAnthropic | closed | effort max; Reasoning effort: max | 77.0%±4.2 | not dated | vals.ai |
| 17 | tencent/Hy4-previewEpoch: hy4-preview_unknown, Tencent | open | effort unknown | 75.0% | not dated | vals.ai |
| 18 | gpt-5.6-terra_xhighOpenAI | closed | effort xhigh; Reasoning effort: xhigh | 74.0%±4.4 | not dated | vals.ai |
| 19 | mimo-v2.6-proXiaomi Corpno page match | open | default | 70.0% | not dated | vals.ai |
| 20 | claude-opus-4-8_maxAnthropic | closed | effort max; Reasoning effort: max | 69.0%±4.6 | not dated | vals.ai |
| 21 | gpt-6-luna_unknownOpenAI | closed | effort unknown | 64.0% | not dated | vals.ai |
| 22 | mimo-v2.6-flashXiaomi Corpno page match | open | default | 63.0% | not dated | vals.ai |
| 23 | gpt-5.6-luna_maxOpenAI | closed | effort max; Reasoning effort: max | 60.0%±4.9 | not dated | vals.ai |
| 24 | gemini-3.7-flash_unknownGoogle DeepMind | closed | effort unknown | 58.0% | not dated | vals.ai |
| 25 | muse-spark-1.3_maxMeta AI | closed | effort max | 58.0% | not dated | vals.ai |
| 26 | qwen3.8-max_unknownAlibaba | closed | effort unknown | 58.0% | not dated | vals.ai |
| 27 | deepseek-ai/DeepSeek-V4-Flash-0731Epoch: deepseek-v4-flash-0731_unknown, DeepSeek | open | effort unknown | 56.0%±5.0 | not dated | vals.ai |
| 28 | gpt-5.4-2026-03-05_xhighOpenAI | closed | effort xhigh; Reasoning effort: xhigh | 56.0%±5.0 | not dated | vals.ai |
| 29 | muse-spark-1.3_unknownMeta AI | closed | effort unknown | 55.0% | not dated | vals.ai |
| 30 | claude-opus-4-7_maxAnthropic | closed | effort max; Reasoning effort: max | 54.0%±5.0 | not dated | vals.ai |
| 31 | deepseek-ai/DeepSeek-V4.1-FlashEpoch: deepseek-v4.1-flash_unknown, DeepSeek | open | effort unknown | 54.0% | not dated | vals.ai |
| 32 | grok-4.6_unknownxAI | closed | effort unknown | 51.0%±5.0 | not dated | vals.ai |
| 33 | claude-opus-4-6_maxAnthropic | closed | effort max; Reasoning effort: max | 50.0%±5.0 | not dated | vals.ai |
| 34 | deepseek-ai/DeepSeek-V4-Pro-0813Epoch: deepseek-v4-pro-0813_unknown, DeepSeek | open | effort unknown | 50.0%±5.0 | not dated | vals.ai |
| 35 | gpt-5.5_xhighOpenAI | closed | effort xhigh; Reasoning effort: xhigh | 50.0%±5.0 | not dated | vals.ai |
| 36 | zai-org/GLM-5.3Epoch: glm-5.3_max, Z.ai (Zhipu AI) | open | effort max | 49.0% | not dated | vals.ai |
| 37 | gemini-3.8-flash_unknownGoogle DeepMind | closed | effort unknown | 48.0% | not dated | vals.ai |
| 38 | claude-sonnet-4-6_maxAnthropic | closed | effort max; Reasoning effort: max | 45.0%±5.0 | not dated | vals.ai |
| 39 | muse-spark-1.2Meta AI | closed | default | 43.0% | not dated | vals.ai |
| 40 | step-5-preview_unknownStepFun | closed | effort unknown | 42.0% | not dated | vals.ai |
| 41 | muse-spark-1.1Meta AI | closed | Reasoning effort: xhigh | 39.0%±4.9 | not dated | vals.ai |
| 42 | claude-opus-4-5-20251101_unknownAnthropic | closed | effort unknown; Reasoning effort: high | 36.0%±4.8 | not dated | vals.ai |
| 43 | gemini-3.6-flash_unknownGoogle DeepMind | closed | effort unknown; Reasoning effort: high | 36.0%±4.8 | not dated | vals.ai |
| 44 | zai-org/GLM-5.2Epoch: glm-5.2_max, Z.ai (Zhipu AI) | open | effort max; Reasoning effort: max | 35.0%±4.8 | not dated | vals.ai |
| 45 | gemini-3.5-flash_highGoogle DeepMind | closed | effort high; Reasoning effort: high | 31.0%±4.6 | not dated | vals.ai |
| 46 | grok-4.5_highxAI | closed | effort high; Reasoning effort: high | 31.0%±4.6 | not dated | vals.ai |
| 47 | gemini-3.1-pro-previewGoogle DeepMind | closed | Reasoning effort: high | 26.0%±4.4 | not dated | vals.ai |
| 48 | grok-4.7_unknownxAI | closed | effort unknown | 26.0% | not dated | vals.ai |
| 49 | qwen3.7-maxAlibaba | closed | default | 26.0%±4.4 | not dated | vals.ai |
| 50 | zai-org/GLM-5.1Epoch: glm-5.1, Z.ai (Zhipu AI) | open | default | 22.2%±4.2 | not dated | vals.ai |
| 51 | XiaomiMiMo/MiMo-V2.5-ProEpoch: mimo-v2.5-pro, Xiaomi Corp | open | default | 22.0%±4.2 | not dated | vals.ai |
| 52 | zai-org/GLM-5.3-FlashEpoch: glm-5.3-flash_max, Z.ai (Zhipu AI) | open | effort max | 21.0%±4.1 | not dated | vals.ai |
| 53 | gpt-5.4-mini-2026-03-17_xhighOpenAI | closed | effort xhigh; Reasoning effort: xhigh | 21.0%±4.1 | not dated | vals.ai |
| 54 | gemini-3-pro-previewGoogle DeepMind | closed | Reasoning effort: high | 20.0%±4.0 | not dated | vals.ai |
| 55 | claude-sonnet-4-5-20250929_unknownAnthropic | closed | effort unknown | 19.0%±3.9 | not dated | vals.ai |
| 56 | beam-1.0_unknownReflectionno page match | open | effort unknown | 18.0%±3.9 | not dated | vals.ai |
| 57 | gpt-5-2025-08-07_highOpenAI | closed | effort high; Reasoning effort: high | 18.0%±3.8 | not dated | vals.ai |
| 58 | MiniMaxAI/MiniMax-M3Epoch: MiniMax-M3, MiniMax | open | default | 18.0%±3.9 | not dated | vals.ai |
| 59 | muse-sparkMeta AI | closed | default | 17.0%±3.8 | not dated | vals.ai |
| 60 | deepseek-ai/DeepSeek-V4-ProEpoch: deepseek-v4-pro_max, DeepSeek | open | effort max; Reasoning effort: max | 16.0%±3.7 | not dated | vals.ai |
| 61 | moonshotai/Kimi-K2.6Epoch: kimi-k2.6, Moonshot | open | default | 16.0%±3.7 | not dated | vals.ai |
| 62 | XiaomiMiMo/MiMo-V2.5Epoch: mimo-v2.5, Xiaomi Corp | open | default | 16.0%±3.7 | not dated | vals.ai |
| 63 | Qwen/Qwen3.8-27BEpoch: qwen3.8-27b, Alibaba | open | default | 16.0% | not dated | vals.ai |
| 64 | gemini-3-flash-previewGoogle DeepMind | closed | Reasoning effort: high | 15.0%±3.6 | not dated | vals.ai |
| 65 | gpt-5.2-2025-12-11_xhighOpenAI | closed | effort xhigh; Reasoning effort: xhigh | 15.0%±3.6 | not dated | vals.ai |
| 66 | grok-4.20-0309-reasoningxAI | closed | default | 14.0%±3.5 | not dated | vals.ai |
| 67 | gemini-3.5-flash-liteGoogle DeepMind | closed | Reasoning effort: high | 13.0%±3.4 | not dated | vals.ai |
| 68 | gpt-5-nano-2025-08-07_highOpenAI | closed | effort high; Reasoning effort: high | 12.0%±3.3 | not dated | vals.ai |
| 69 | grok-4.3_highxAI | closed | effort high; Reasoning effort: high | 11.0%±3.1 | not dated | vals.ai |
| 70 | mistral-large-4-0_unknownMistral AIno page match | open | effort unknown | 10.0%±3.0 | not dated | vals.ai |
| 71 | gpt-5-mini-2025-08-07_highOpenAI | closed | effort high; Reasoning effort: high | 9.0%±2.9 | not dated | vals.ai |
| 72 | gpt-5.1-codex-maxOpenAI | closed | Reasoning effort: high | 9.0%±2.9 | not dated | vals.ai |
| 73 | mistral-medium-2604Mistral AIno page match | open | Reasoning effort: high | 9.0%±2.9 | not dated | vals.ai |
| 74 | deepseek-ai/DeepSeek-V3.2Epoch: fireworks/deepseek-v3p2, DeepSeek | open | Reasoning effort: none | 8.0%±2.7 | not dated | vals.ai |
| 75 | zai-org/GLM-4.7Epoch: glm-4.7, Z.ai (Zhipu AI) | open | default | 6.0%±2.4 | not dated | vals.ai |
| 76 | thinkingmachines/Inkling-SmallEpoch: inkling-small, Thinking Machines | open | Reasoning effort: 0.99 | 6.0%±2.4 | not dated | vals.ai |
| 77 | gpt-5.4-nano-2026-03-17_highOpenAI | closed | effort high; Reasoning effort: high | 5.0%±2.2 | not dated | vals.ai |
| 78 | grok-4-1-fast-reasoningxAI | closed | default | 4.0%±2.0 | not dated | vals.ai |
| 79 | MiniMaxAI/MiniMax-M2.5Epoch: MiniMax-M2.5, MiniMax | open | default | 4.0%±2.0 | not dated | vals.ai |
| 80 | mercury-2.5_unknownInception Labs | closed | effort unknown | 3.0% | not dated | vals.ai |
| 81 | MiniMaxAI/MiniMax-M2.7Epoch: MiniMax-M2.7, MiniMax | open | default | 3.0%±1.7 | not dated | vals.ai |
| 82 | nemotron-3-ultraNvidiano page match | open | default | 2.0%±1.4 | not dated | vals.ai |
| 83 | thinkingmachines/InklingEpoch: Inkling, Thinking Machines | open | default | 0.0% | not dated | vals.ai |
| 84 | poolside/Laguna-M.1Epoch: laguna-m.1, Poolside | open | default | 0.0% | not dated | vals.ai |
| 85 | poolside/Laguna-XS.2Epoch: laguna-xs.2, Poolside | open | default | 0.0% | not dated | vals.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