Benchmark
Humanity's Last Exam
Scores are percentages. Higher is better.
- Models and systems measured
- 120
- Scores recorded
- 230
- Sources
- 3
- IQ weight
- 1.67
Score distribution
- 34%
- Mean
- 34.8%
- Median
- 15.5%
- Std dev
- 12.5%
- 10th percentile
- 55.7%
- 90th percentile
- 3.7%
- Min
This benchmark has a weight of 1.67 for chat models, measured over 113 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 |
|---|---|---|
| 51 | Claude Opus 4.7no tools · max effort · ≤1M total tokens · no compaction· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 46.9% |
| 52 | Claude Opus 4.7Opus 4.8 System Card capability summary · adaptive thinking · max effort · 5 trials unless noted · no tools· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 46.9% |
| 53 | Claude Haiku 5.5no tools · 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. | 45.9% |
| 54 | Gemini 3 Pro Preview· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 45.8% |
| 55 | Claude Opus 4.8Qwen表:无工具· Reported by Qwen· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 45.7% |
| 56 | Gemini 3.1 Pro Previewhigh · no tools · full 2,500 questions · o3-mini judge· Reported by Meta· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 45.4% |
| 57 | GPT 5.5xhigh · no tools · full 2,500 questions · o3-mini judge· Reported by Meta· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 44.8% |
| 58 | GPT-5.6 SolDeepSeek V4.1 launch chart · exact public value · setting not further specified · full set· Reported by DeepSeek· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 44.5% |
| 59 | Gemini 3.1 Pro PreviewFable/Mythos 5 System Card capability summary · adaptive thinking · max effort · 5 trials unless noted · no tools· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 44.4% |
| 60 | Gemini 3.1 Pro PreviewGemini 3.5 Flash Model Card · May 2026 · full text + multimodal set· Reported by Google DeepMind· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 44.4% |
| 61 | Gemini 3.1 Pro Previewhigh · no tools · self-reported value quoted beside Meta's 45.4 reproduction· Reported by Meta· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 44.4% |
| 62 | Gemini 3.1 Pro Previewno tools· Reported by Google DeepMind· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 44.4% |
| 63 | Gemini 3.1 Pro PreviewOpus 4.8 System Card capability summary · adaptive thinking · max effort · 5 trials unless noted · no tools· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 44.4% |
| 64 | Inkling 2 Small· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 44.1% |
| 65 | DeepSeek V4.1 Flash 1017DeepSeek 1017 first-party evaluation · reasoning effort, candidate sampling, exact run count, and unspecified harnesses not published · text-only HLE subset · no tools· Reported by DeepSeek· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 44% |
| 66 | Qwen3.8 Max· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 43.6% |
| 67 | Qwen3.8 MaxQwen表:无工具· Reported by Qwen· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 43.6% |
| 68 | Gemini 3 Flash· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 43.5% |
| 69 | Kimi K3DeepSeek V4.1 launch chart · exact public value · setting not further specified · full set· Reported by DeepSeek· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 43.5% |
| 70 | Kimi K3full set · no tools · max effort · temp 1.0 · top_p .95· Reported by Moonshot AI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 43.5% |
| 71 | Hy4 previewtext-only subset · no tools· Reported by Tencent Hy· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 43.4% |
| 72 | MiniMax M3.1 FlashMiniMax M3.1 Flash official-weights model-card run · reasoning effort, sampling, and run count not stated for this benchmark · without tools · text-only subset / 2,158 questions· Reported by MiniMax· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 43.4% |
| 73 | Claude Sonnet 5no tools · auto thinking · 1M total tokens· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 43.2% |
| 74 | GPT-5.5 Prono tools · xhigh· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 43.1% |
| 75 | GLM 4.7· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claimValues from 2 sources
| 42.8% |
| 76 | DeepSeek V4 Pro 0813DeepSeek V4.1 launch chart · exact public value · setting not further specified · text-only subset· Reported by DeepSeek· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 42.7% |
| 77 | DeepSeek V4 Pro 0813no tools · pass@1· Reported by DeepSeek· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 42.7% |
| 78 | GPT-5.4 Prono tools· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 42.7% |
| 79 | DeepSeek V4.1 Flash 1017· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 42.6% |
| 80 | DeepSeek V4.1 Flash 1017DeepSeek 1017 first-party evaluation · reasoning effort, candidate sampling, exact run count, and unspecified harnesses not published · full HLE set · no tools· Reported by DeepSeek· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 42.6% |
| 81 | GLM 5.3DeepSeek V4.1 launch chart · exact public value · setting not further specified · text-only subset· Reported by DeepSeek· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 42% |
| 82 | GPT 5.5Fable/Mythos 5 System Card capability summary · adaptive thinking · max effort · 5 trials unless noted · no tools· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 41.4% |
| 83 | GPT 5.5Gemini 3.5 Flash Model Card · May 2026 · full text + multimodal set· Reported by Google DeepMind· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 41.4% |
| 84 | GPT 5.5no tools · xhigh· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 41.4% |
| 85 | GPT 5.5Opus 4.8 System Card capability summary · adaptive thinking · max effort · 5 trials unless noted · no tools· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 41.4% |
| 86 | GPT 5.5xhigh · no tools · self-reported value quoted beside Meta's 44.8 reproduction· Reported by Meta· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 41.4% |
| 87 | Qwen3.7 Max· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 41.4% |
| 88 | Qwen3.7 MaxQwen3.7-Max release · recommended reasoning-effort=xhigh system prompt· Reported by Qwen· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 41.4% |
| 89 | Qwen3.7 MaxQwen表:无工具· Reported by Qwen· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 41.4% |
| 90 | DeepSeek V3.2· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claimValues from 2 sources
| 40.8% |
| 91 | GLM 5.2text-only subset · temperature=1 · top_p=.95 · max generation 163,840· Reported by Z.ai· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 40.5% |
| 92 | DeepSeek V4 Flash· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 40.3% |
| 93 | Gemini 3.5 Flash· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 40.2% |
| 94 | Gemini 3.5 FlashGemini 3.5 Flash Model Card · May 2026 · full text + multimodal set· Reported by Google DeepMind· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 40.2% |
| 95 | Claude Opus 4.6Anthropic Opus 4.6 launch page · adaptive thinking · max effort unless noted · no tools· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 40% |
| 96 | Claude Opus 4.6GPT-4o judge · max effort· Reported by Qwen· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 40% |
| 97 | Grok 4· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 40% |
| 98 | GPT 5.4· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 39.8% |
| 99 | GPT 5.4no tools· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 39.8% |
| 100 | DeepSeek V4.1 FlashDeepSeek V4.1 launch chart · exact public value · setting not further specified · text-only subset· Reported by DeepSeek· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 39.1% |
Showing 51–100 of 229 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 · no tools · full 2,500 questions · o3-mini judge· Reported by Meta· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 42.8% |
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. · OpenEvals/leaderboard-data, MIT (licence)