Benchmark
Humanity's Last Exam (with tools)
Scores are percentages. Higher is better.
- Models and systems measured
- 61
- Scores recorded
- 113
- Sources
- 2
- IQ weight
- 1.73
Score distribution
- 51.4%
- Mean
- 54%
- Median
- 13.5%
- Std dev
- 36.5%
- 10th percentile
- 63.9%
- 90th percentile
- 6.3%
- Min
This benchmark has a weight of 1.73 for chat models, measured over 14 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.
Leaderboard
| # | Model or system | Score |
|---|---|---|
| 1 | Claude Fable 5.2· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 74.2% |
| 2 | Kimi K3.1 Pro· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 70% |
| 3 | Claude Opus 5.5Opus 5.5 System Card capability section · web search/fetch + programmatic tools + code · auto thinking · 1M total-token cap· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 67.7% |
| 4 | Claude Fable 5.1Opus 5.5 System Card capability section · with tools· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 65.6% |
| 5 | Claude Fable 5.1System Card capability summary · adaptive thinking · max effort · default sampling · 5 trials · context ≤1M unless noted · with tools· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 65% |
| 6 | Claude Fable 5.1with tools· Reported by OpenAI, Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 65% |
| 7 | Claude Opus 5Opus 5 System Card capability summary · adaptive thinking · max effort unless noted · with tools· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 64.7% |
| 8 | Claude Fable 5Qwen表:with tools· Reported by Qwen· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 64.5% |
| 9 | Claude Sonnet 5.5· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 64.5% |
| 10 | Claude Sonnet 5.5web search/fetch + programmatic tools + code · auto thinking · 980K task budget· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 64.5% |
| 11 | GPT-5.6 SolZ.ai设置· Reported by Z.ai· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 64.5% |
| 12 | Claude Fable 5Opus 5 System Card capability summary · adaptive thinking · max effort unless noted · with tools· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 63.9% |
| 13 | DeepSeek V4.1 Flash· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 63.9% |
| 14 | DeepSeek V4.1 FlashDeepSeek V4.1 launch chart · exact public value · setting not further specified· Reported by DeepSeek· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 63.9% |
| 15 | DeepSeek V4.1 FlashPass@1 / tools· Reported by DeepSeek· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 63.9% |
| 16 | Claude Fable 5System Card capability summary · adaptive thinking · max effort · default sampling · 5 trials · context ≤1M unless noted · with tools· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 63.8% |
| 17 | Claude Fable 5with tools· Reported by OpenAI, Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 63.8% |
| 18 | Claude Opus 5DeepSeek V4.1 launch chart · exact public value · setting not further specified· Reported by DeepSeek· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 63.6% |
| 19 | Claude Opus 5Opus 5.5 System Card capability section · with tools· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 63.6% |
| 20 | Claude Opus 5System Card capability summary · adaptive thinking · max effort · default sampling · 5 trials · context ≤1M unless noted · with tools· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 63.6% |
| 21 | Claude Opus 5with tools· Reported by OpenAI, Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 63.6% |
| 22 | GLM 5.3DeepSeek V4.1 launch chart · exact public value · setting not further specified· Reported by DeepSeek· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 62.5% |
| 23 | GLM 5.3Z.ai设置· Reported by Z.ai· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 62.5% |
| 24 | Muse Spark 1.1Meta Model API · xhigh · full 2,500 questions · bash + browser tools · o3-mini judge· Reported by Meta, Meta Superintelligence Labs· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 62.1% |
| 25 | Muse Spark Open· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 61% |
| 26 | DeepSeek V4 Pro 0813DeepSeek V4.1 launch chart · exact public value · setting not further specified· Reported by DeepSeek· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 60% |
| 27 | DeepSeek V4 Pro 0813with tools · pass@1· Reported by DeepSeek· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 60% |
| 28 | DeepSeek V4 Pro 0813Z.ai设置· Reported by Z.ai· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 60% |
| 29 | Kimi K3DeepSeek V4.1 launch chart · exact public value · setting not further specified· Reported by DeepSeek· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 59.8% |
| 30 | Kimi K3Z.ai设置· Reported by Z.ai· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 59.8% |
| 31 | GPT-5.4 Prowith tools· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 58.7% |
| 32 | GPT-5.6 SolQwen表:with tools· Reported by Qwen· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 58% |
| 33 | Claude Opus 4.8Fable/Mythos 5 System Card capability summary · adaptive thinking · max effort · 5 trials unless noted · with tools· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 57.9% |
| 34 | Claude Opus 4.8max · full 2,500 questions · bash + browser tools · o3-mini judge· Reported by Meta, Meta Superintelligence Labs· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 57.9% |
| 35 | Claude Opus 4.8Opus 4.8 System Card capability summary · adaptive thinking · max effort · 5 trials unless noted · with tools· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 57.9% |
| 36 | Claude Opus 4.8Opus 5 System Card capability summary · adaptive thinking · max effort unless noted · with tools· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 57.9% |
| 37 | Claude Opus 4.8Qwen表:with tools· Reported by Qwen· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 57.9% |
| 38 | Claude Opus 4.8Z.ai设置· Reported by Z.ai· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 57.9% |
| 39 | Claude Haiku 5.5· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 57.4% |
| 40 | Claude Haiku 5.5web search/fetch + programmatic tools + code · auto thinking · 980K task budget· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 57.4% |
| 41 | Claude Sonnet 5search/fetch/code · max effort · 1M total tokens· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 57.4% |
| 42 | GPT-5.5 Prowith tools · xhigh· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 57.2% |
| 43 | GPT-6 AstraOpus 5.5 System Card capability section · with tools · provider-reported· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 57.2% |
| 44 | GPT-6 Astrawith tools· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 57.2% |
| 45 | Ling 3.0 Flash SanteLing-3.0-flash-Sante official evaluation image · highest reasoning tier · temperature 1.0 / top_p 0.95 unless otherwise stated · exact API snapshot and run count not published · with tools · 256K · no context-management strategy· Reported by Ant inclusionAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 57.1% |
| 46 | Qwen3.8 MaxQwen表:with tools· Reported by Qwen· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 56.2% |
| 47 | Qwen3.8 MaxZ.ai设置· Reported by Z.ai· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 56.2% |
| 48 | Kimi K3full set · tools · max effort · temp 1.0· Reported by Moonshot AI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 56% |
| 49 | Kimi K2.6Kimi K2.6 official benchmark table · thinking · temp 1.0 · top_p 1.0 · 262,144 context · text-only subset · with tools · footnote result· Reported by Moonshot AI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 55.5% |
| 50 | Hy4 previewtext-only subset · with tools· Reported by Tencent Hy· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 55.4% |
Showing 1–50 of 112 results.
Not ranked
The source marks these results as descriptive, not capability scores, so we list them by name instead of ranking them.
| Model or system | Score |
|---|---|
| Muse Spark (version unspecified)Meta Model API · highest effort · version unspecified · full 2,500 questions · bash + browser tools · o3-mini judge· Reported by Meta, Meta Superintelligence Labs· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 50.4% |
Data credits: Data by LLM Stats · BenchBoard by SYuan03 (github.com/SYuan03/benchboard). MIT covers its code and data organization; no results licence is inferred. Numeric relays retain the original stating-source links.