Vals ProofBench (ProofBench)
We show this table for reference; we do not rank on it.
Vals AI automated theorem-proving benchmark.
Evaluation results published by Vals AI. Published leaderboard aggregates and evaluation settings only. Dataset contents are not included.
ProofBench score on ProofBench — November 9, 2026
We mirror the published proofbench score view for ProofBench. Claude Fable 5.2 leads the public snapshot at 100.00%, followed by Claude Sonnet 5.5 (100.00%) and Claude Opus 5.5 (100.00%). We do not use these results to rank models overall.
Claude Fable 5.2
Anthropic
Claude Sonnet 5.5
Anthropic
Claude Opus 5.5
Anthropic
55 modelsMathematicsCurrentDisplay onlyUpdated November 9, 2026
ProofBench score table (55 models)
| Rank | Model / configuration | Score | Parameters (B) | Open / closed |
|---|---|---|---|---|
| 1 | Claude Fable 5.2Anthropic | 100.00% | Not reported | Closed |
| 2 | Claude Sonnet 5.5Anthropic | 100.00% | Not reported | Closed |
| 3 | Claude Opus 5.5Anthropic | 100.00% | Not reported | Closed |
| 4 | Claude Fable 5.1Anthropic | 100.00% | Not reported | Closed |
| 5 | AlephproverLogical Intelligence | 100.00% | Not reported | Unknown |
| 6 | GPT-6.1 SolOpenAImax reasoning | 99.00% | Not reported | Closed |
| 7 | GPT-6 AstraOpenAImax reasoning | 99.00% | Not reported | Closed |
| 8 | Gemini 4 ArgonGooglehigh reasoning | 99.00% | Not reported | Closed |
| 9 | Claude Opus 5Anthropic | 99.00% | Not reported | Closed |
| 10 | Claude Fable 5Anthropic | 95.00% | Not reported | Closed |
| 11 | Muse Wonder FullMetamax reasoning | 94.00% | Not reported | Closed |
| 12 | Kimi K3.1 ProMoonshot AImax reasoning | 91.00% | Not reported | Open |
| 13 | Claude Haiku 5.5Anthropic | 89.00% | Not reported | Closed |
| 14 | Kimi K3Moonshot AImax reasoning | 87.00% | Not reported | Open |
| 15 | AristotleHarmonic | 86.00% | Not reported | Unknown |
| 16 | Muse WonderMetamax reasoning | 84.00% | Not reported | Closed |
| 17 | GPT-6 SolOpenAImax reasoning | 83.00% | Not reported | Closed |
| 18 | GPT-5.6 SolOpenAImax reasoning | 83.00% | Not reported | Closed |
| 19 | Claude Sonnet 5Anthropic | 77.00% | Not reported | Closed |
| 20 | Hy4 previewTencent | 75.00% | Not reported | Open |
| 21 | GPT-5.6 TerraOpenAImax reasoning | 74.00% | Not reported | Closed |
| 22 | MiMo-V2.6-ProXiaomi | 70.00% | Not reported | Open |
| 23 | GPT-6 LunaOpenAImax reasoning | 64.00% | Not reported | Closed |
| 24 | MiMo-V2.6-FlashXiaomi | 63.00% | Not reported | Open |
| 25 | Ember-1Fireworks | 61.00% | Not reported | Closed |
| 26 | GPT-5.6 LunaOpenAImax reasoning | 60.00% | Not reported | Closed |
| 27 | Gemini 3.7 FlashGooglehigh reasoning | 58.00% | Not reported | Closed |
| 28 | Muse Spark 1.3 MaxMetamax reasoning | 58.00% | Not reported | Unknown |
| 29 | Qwen3.8 MaxAlibaba | 58.00% | Not reported | Open |
| 30 | DeepSeek V4 Flash 0731DeepSeekhigh reasoning | 56.00% | Not reported | Open |
| 31 | Muse Spark 1.3Metaxhigh reasoning | 55.00% | Not reported | Closed |
| 32 | DeepSeek V4.1 Flash 1017DeepSeekhigh reasoning | 54.00% | Not reported | Open |
| 33 | Grok 4.6xAIhigh reasoning | 51.00% | Not reported | Closed |
| 34 | DeepSeek V4 Pro 0813DeepSeekmax reasoning | 50.00% | Not reported | Open |
| 35 | GLM-5.3Z.AImax reasoning | 49.00% | Not reported | Open |
| 36 | Gemini 3.8 FlashGooglehigh reasoning | 48.00% | Not reported | Closed |
| 37 | Muse Spark 1.2Metaxhigh reasoning | 43.00% | Not reported | Closed |
| 38 | Step 5 PreviewStepFun | 42.00% | Not reported | Open |
| 39 | Gemini 3.5 FlashGooglehigh reasoning | 31.00% | Not reported | Closed |
| 40 | Grok 4.5xAIhigh reasoning | 31.00% | Not reported | Closed |
| 41 | Grok 4.7xAIxhigh reasoning | 26.00% | Not reported | Closed |
| 42 | Gemini 3.1 Pro PreviewGooglehigh reasoning | 26.00% | Not reported | Unknown |
| 43 | MiMo-V2.5-ProXiaomi | 22.00% | Not reported | Closed |
| 44 | GLM-5.3-FlashZ.AImax reasoning | 21.00% | Not reported | Open |
| 45 | Beam 1.0Reflection | 18.00% | Not reported | Unknown |
| 46 | MiniMax M3MiniMax | 18.00% | Not reported | Open |
| 47 | DeepSeek V4 Pro 0813DeepSeekmax reasoning | 16.00% | Not reported | Open |
| 48 | MiMo-V2.5Xiaomi | 16.00% | Not reported | Closed |
| 49 | Qwen3.8-27BAlibabaxhigh reasoning | 16.00% | Not reported | Open |
| 50 | Mistral Large 4Mistralhigh reasoning | 10.00% | Not reported | Open |
| 51 | Mistral Medium 3.5Mistral AIhigh reasoning | 9.00% | Not reported | Unknown |
| 52 | Inkling-SmallThinking Machines Lab0.99 reasoning | 6.00% | Not reported | Open |
| 53 | Mercury 2.5Inceptionhigh reasoning | 3.00% | Not reported | Closed |
| 54 | InklingThinking Machines Lab0.99 reasoning | 0.00% | Not reported | Open |
| 55 | Nemotron Lightning 3p5 30b A3bFireworks AI | 0.00% | Not reported | Unknown |
How ProofBench is shown here
BenchLM mirrors the public Vals AI ProofBench leaderboard captured from https://www.vals.ai/benchmarks/proof_bench and updated by Vals on November 9, 2026. The snapshot preserves overall scores, uncertainty, latency, cost-per-test metadata, and task-level scores where Vals publishes them.
ProofBench is display only on BenchLM. Vals proprietary or Vals-hosted aggregate views are useful context, but BenchLM does not use them as weighted ranking inputs or as a replacement for benchmark-native source records.
Snapshot
The published ProofBench snapshot places Claude Fable 5.2 first at 100.00%. The third row is 0.00 points behind. The broader top-10 range is 5.00 points, so many of the published results sit in a relatively narrow band.
55 models have been evaluated on ProofBench. The benchmark falls in the Mathematics category. We keep external benchmark mirrors separate from the weighted global scoring system, so these results remain source-specific evidence. ProofBench is currently displayed for reference but excluded from the scoring formula, so it does not directly affect overall rankings.
About ProofBench
Year
2026
Tasks
Automated theorem proving
Format
Accuracy score
Difficulty
Formal proof reasoning
BenchLM mirrors Vals ProofBench as a display-only math and proof benchmark.
Freshness and provenance
Version
ProofBench 2026
Refresh cadence
Quarterly
Staleness state
Current
Question availability
Public benchmark set
BenchLM uses freshness metadata to decide whether a benchmark should still be treated as a strong differentiator, a benchmark to watch, or a display-only reference. For the full scoring policy, see the BenchLM methodology page.
Questions
What does ProofBench measure?
Vals AI automated theorem-proving benchmark.
Which model leads the published ProofBench snapshot?
Claude Fable 5.2 currently leads the published ProofBench snapshot with 100.00% proofbench score. BenchLM shows this benchmark for display only and does not use it in overall rankings.
How many models are evaluated on ProofBench?
The November 9, 2026 snapshot contains 55 AI models.
Know when it’s worth switching models
The model to choose, the cheaper alternative, and the release we would wait on.
Read a sample issueJoin 7,000+ subscribers.
One email each week. Unsubscribe anytime.