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 |
|---|---|---|
| 51 | GLM 5.3 Flashfull set / tools· Reported by Z.ai· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 55.3% |
| 52 | Hy4 Flash· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 55.2% |
| 53 | Claude Opus 4.7Opus 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. | 54.7% |
| 54 | Claude Opus 4.7web search/fetch + code · 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. | 54.7% |
| 55 | GLM 5.2tools · 300K context · no context management· Reported by Z.ai· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 54.7% |
| 56 | Claude Sonnet 5search/fetch/code · xhigh effort · 1M total tokens· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 54.6% |
| 57 | Kimi K2.6Kimi K2.6 official benchmark table · thinking · temp 1.0 · top_p 1.0 · 262,144 context · full set · search + code interpreter + browser · 262,144 max generation · context management· Reported by Moonshot AI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 54% |
| 58 | Qwen3.7 MaxQwen3.7-Max release · published setting· Reported by Qwen· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 53.5% |
| 59 | Qwen3.7 MaxQwen表:with tools· Reported by Qwen· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 53.5% |
| 60 | Hy3 (Deep Infra)Hy3 official Benchmark Appendix · highest reasoning tier unless noted · text-only · with tools· Reported by Tencent Hunyuan· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 53.2% |
| 61 | Claude Opus 4.6Anthropic Opus 4.6 launch page · adaptive thinking · max effort unless noted · with tools · original chart label· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 53.1% |
| 62 | Claude Opus 4.6Anthropic Opus 4.6 launch page · adaptive thinking · max effort unless noted · with tools · revised 2026-02-23· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 53% |
| 63 | Claude Sonnet 5search/fetch/code · high effort · 1M total tokens· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 52.8% |
| 64 | GLM 5.1GLM-5.1 official release table· Reported by Z.ai· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 52.3% |
| 65 | GPT 5.5Fable/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. | 52.2% |
| 66 | GPT 5.5Opus 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. | 52.2% |
| 67 | GPT 5.5with tools · xhigh· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 52.2% |
| 68 | GPT 5.5xhigh · 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. | 52.2% |
| 69 | GPT 5.4with tools· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 52.1% |
| 70 | Kimi K2.5text subset · tools· Reported by Moonshot AI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 51.8% |
| 71 | DeepSeek V4 Flash 0731DeepSeek 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. | 51.5% |
| 72 | Gemini 3.1 Pro PreviewFable/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. | 51.4% |
| 73 | Gemini 3.1 Pro Previewhigh · 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. | 51.4% |
| 74 | Gemini 3.1 Pro PreviewOpus 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. | 51.4% |
| 75 | Gemini 3.1 Pro Previewsearch with blocklist + code· Reported by Google DeepMind· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 51.4% |
| 76 | Beam 1.0 501B A23B· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 51.1% |
| 77 | Qwen3.6 PlusQwen official release table · thinking mode unless benchmark requires otherwise · 256K · context folding· Reported by Alibaba Qwen· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 50.6% |
| 78 | GLM 5GLM-5 official release table· Reported by Z.ai· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 50.4% |
| 79 | Kimi K2.5full set · tools· Reported by Moonshot AI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 50.2% |
| 80 | GPT-5.2 Prosearch + Python · xhigh· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 50% |
| 81 | GPT-5.2 Prowith tools· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 50% |
| 82 | Claude Sonnet 4.6web search + web fetch + code · 3M total tokens · adaptive thinking· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 49% |
| 83 | Qwen3.5 397B A17BQwen3.5 release/model card · published setting· Reported by Qwen· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 48.3% |
| 84 | Step 3.7 FlashStep 3.7 Flash official README/figure · with tools · prose value· Reported by StepFun· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 48.1% |
| 85 | MiMo V2.5 ProXiaomi MiMo-V2.5-Pro official launch table · with tools· Reported by Xiaomi MiMo· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 48% |
| 86 | Claude Sonnet 5search/fetch/code · medium effort · 1M total tokens· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 47.2% |
| 87 | Step 3.7 FlashStep 3.7 Flash official README/figure · with tools · figure value· Reported by StepFun· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 47.2% |
| 88 | Claude Sonnet 4.6with tools · max effort · Sonnet 5 card rerun· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 46.8% |
| 89 | Gemini 3 Pro PreviewGemini 3 Pro Model Card · results as of November 2025 · search + code execution· Reported by Google DeepMind· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 45.8% |
| 90 | GPT 5.2search + Python · xhigh· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 45.5% |
| 91 | GPT 5.2with tools· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 45.5% |
| 92 | DeepSeek V4 FlashDeepSeek V4 official model card · Think Max· Reported by DeepSeek· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 45.1% |
| 93 | Gemini 3 Flashweb search + code execution · pass@1· Reported by Google DeepMind· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 43.5% |
| 94 | Claude Opus 4.5web search/fetch + code · no reasoning · decontaminated · Sonnet 4.5 grader· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 43.2% |
| 95 | GPT-5.4 minixhigh · with tool· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 41.5% |
| 96 | DeepSeek V3.2Thinking mode · temperature 1.0 · 128K context · Search Agent · commercial search API· Reported by DeepSeek· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 40.8% |
| 97 | DeepSeek V4 FlashDeepSeek V4 official model card · Think High· Reported by DeepSeek· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 40.3% |
| 98 | Kimi K2.5image subset · tools· Reported by Moonshot AI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 39.8% |
| 99 | GPT 5.4 Nanoxhigh · with tool· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 37.7% |
| 100 | Nemotron 3 UltraNeMo Evaluator SDK · with tools· Reported by NVIDIA· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 37.4% |
Showing 51–100 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.