llmrun
Math

ProofBench Leaderboard

ProofBench evaluates a model's ability to write and check rigorous mathematical proofs, rather than just produce a final numeric answer. It probes step-by-step logical correctness, which is a harder and more distinct skill than solving for a number.

Source: epoch21 open models ranked+43 proprietaryData through Sep 2026

Open modelsAll models

All models ranked on ProofBench

Proprietary / closed models are shown dimmed — you can't run them locally, but they show where the open field stands.

#ModelScore
1Claude Fable 5.1 Max · proprietary
100.0%
2Claude Opus 5 Max · proprietary
99.0%
3GPT 6 Astra (unspecified) · proprietary
99.0%
4Claude Fable 5 Max · proprietary
95.0%
5Kimi K3 · 2779.9B
87.0%
6GPT 5.6 Sol Max · proprietary
83.0%
7Claude Sonnet 5 Max · proprietary
77.0%
8GPT 5.6 Terra (xhigh) · proprietary
74.0%
9Claude Opus 4.8 Max · proprietary
69.0%
10GPT 5.6 Luna Max · proprietary
60.0%
11Gemini 3.7 Flash (unspecified) · proprietary
58.0%
12Qwen3.8 Max (unspecified) · proprietary
58.0%
13DeepSeek V4 Flash 0731 · 304.2B
56.0%
14GPT 5.4 (Mar 05, 2026, xhigh) · proprietary
56.0%
15Claude Opus 4.7 Max · proprietary
54.0%
16Grok 4.6 (unspecified) · proprietary
51.0%
17Claude Opus 4.6 Max · proprietary
50.0%
18DeepSeek V4 Pro 0813 · 1650.5B
50.0%
19GPT 5.5 (xhigh) · proprietary
50.0%
20GLM 5.3 · 753.3B
49.0%
21Gemini 3.8 Flash (unspecified) · proprietary
48.0%
22Claude Sonnet 4.6 Max · proprietary
45.0%
23Muse Spark 1.2 · proprietary
43.0%
24Muse Spark 1.1 · proprietary
39.0%
25Claude Opus 4.5 (Nov 01, 2025, unspecified) · proprietary
36.0%
26Gemini 3.6 Flash (unspecified) · proprietary
36.0%
27GLM 5.2 · 753.3B
35.0%
28Gemini 3.5 Flash (high) · proprietary
31.0%
29Grok 4.5 (high) · proprietary
31.0%
30Gemini 3.1 Pro Preview · proprietary
26.0%
31Qwen3.7 Max · proprietary
26.0%
32GLM 5.1 · 753.9B
22.2%
33MiMo V2.5 Pro · 1023.2B
22.0%
34GLM 5.3 Flash · 321.3B
21.0%
35GPT 5.4 Mini (Mar 17, 2026, xhigh) · proprietary
21.0%
36Gemini 3 Pro Preview · proprietary
20.0%
37Claude Sonnet 4.5 (Sep 29, 2025, unspecified) · proprietary
19.0%
38GPT 5 (Aug 07, 2025, high) · proprietary
18.0%
39MiniMax M3 · 427.0B
18.0%
40Muse Spark · proprietary
17.0%
41DeepSeek V4 Pro · 1598.8B
16.0%
42Kimi K2.6 · 1026.9B
16.0%
43MiMo V2.5 · 310.8B
16.0%
44Qwen3.8 27B · 27.8B
16.0%
45Gemini 3 Flash Preview · proprietary
15.0%
46GPT 5.2 (Dec 11, 2025, xhigh) · proprietary
15.0%
47Grok 4.20 0309 Reasoning · proprietary
14.0%
48Gemini 3.5 Flash Lite · proprietary
13.0%
49GPT 5 Nano (Aug 07, 2025, high) · proprietary
12.0%
50Grok 4.3 (high) · proprietary
11.0%
51GPT 5 Mini (Aug 07, 2025, high) · proprietary
9.0%
52GPT 5.1 Codex Max · proprietary
9.0%
53Mistral Medium 2604 · proprietary
9.0%
54DeepSeek V3.2 · 685.4B
8.0%
55GLM 4.7 · 358.3B
6.0%
56Inkling Small · 266.0B
6.0%
57GPT 5.4 Nano (Mar 17, 2026, high) · proprietary
5.0%
58Grok 4.1 Fast Reasoning · proprietary
4.0%
59MiniMax M2.5 · 228.7B
4.0%
60MiniMax M2.7 · 228.7B
3.0%
61Nemotron 3 Ultra · proprietary
2.0%
62Inkling · 952.4B
0.0%
63Laguna M.1 · 225.8B
0.0%
64Laguna XS.2 · 33.4B
0.0%

Score vs model size

Which models give the most quality for their size — the ones worth running locally.

100B1Tmodel size (log scale) →87.0%0.0%DeepSeek V4 Pro 0813 · 1.7T · 50.0%GLM 5.3 · 753B · 49.0%GLM 5.2 · 753B · 35.0%GLM 5.1 · 754B · 22.2%MiMo V2.5 Pro · 1T · 22.0%GLM 5.3 Flash · 321B · 21.0%MiniMax M3 · 427B · 18.0%DeepSeek V4 Pro · 1.6T · 16.0%Kimi K2.6 · 1T · 16.0%MiMo V2.5 · 311B · 16.0%DeepSeek V3.2 · 685B · 8.0%Inkling Small · 266B · 6.0%GLM 4.7 · 358B · 6.0%MiniMax M2.5 · 229B · 4.0%MiniMax M2.7 · 229B · 3.0%Laguna M.1 · 226B · 0.0%Laguna XS.2 · 33B · 0.0%Inkling · 952B · 0.0%Qwen3.8 27B · 28B · 16.0%Qwen3.8 27BDeepSeek V4 Flash 0731 · 304B · 56.0%DeepSeek V4 Flash 0731Kimi K3 · 2.8T · 87.0%Kimi K3
Each dot is a model. Up = higher score, left = smaller (easier to run locally). The dashed line marks the efficiency frontier — the best score you can get at each size or smaller.
  • Qwen3.8 27B, 28B, score 16.0% — on the efficiency frontier (best score at its size or smaller).
  • DeepSeek V4 Flash 0731, 304B, score 56.0% — on the efficiency frontier (best score at its size or smaller).
  • Kimi K3, 2.8T, score 87.0% — on the efficiency frontier (best score at its size or smaller).

ProofBench: frequently asked questions

What is the best open LLM on ProofBench?
Kimi K3 is the top open model on ProofBench, scoring 87.0%. Among all models tested — including proprietary ones — it ranks #5. The top model overall is Claude Fable 5.1 Max (Anthropic) at 100.0%.
What's the best ProofBench model you can run on a 24 GB GPU?
Qwen3.8 27B is the highest-scoring open model that fits in 24 GB at 4-bit quantization (about 15 GB), scoring 16.0% on ProofBench.
Can open models match proprietary models on ProofBench?
Not quite on ProofBench: the strongest proprietary model (Claude Fable 5.1 Max) scores 100.0%, ahead of the best open model (Kimi K3) at 87.0% — but you can run the open one yourself.

Scores aggregated from epoch. llmrun does not run this benchmark — see the source for methodology, or the about benchmarks for what it measures.