Knowledge benchmark
HLE leaderboard
Humanity's Last Exam. Every model the catalog carries a published HLE value for, ranked by that value.
An expert-authored benchmark designed to probe frontier knowledge and reasoning. BenchLM keeps protocol differences visible because tool-assisted and closed-book HLE runs answer different questions.
HLE ranking
58 models with a published HLE value, ordered by that value, highest first. Models the source has not scored on this benchmark are not listed — they are unmeasured, not last.
| Rank | Model | Provider | Open-ended and multiple choice |
|---|---|---|---|
| 1 |  Claude Fable 5.1 | Anthropic | 65 |
| 2 | Claude Opus 5 | Anthropic | 64.7 |
| 3 | Claude Mythos 5 | Anthropic | 64.5 |
| 4 |  Muse Spark 1.1 | Meta | 62.1 |
| 5 |  GPT-5.4 Pro | OpenAI | 58.7 |
| 6 | Claude Opus 4.8 | Anthropic | 57.9 |
| 7 | Claude Sonnet 5 | Anthropic | 57.4 |
| 8 | GPT-5.5 Pro | OpenAI | 57.2 |
| 9 | APApodex 1.1 | Apodex | 56.1 |
| 10 |  Kimi K3 | Moonshot AI | 56 |
| 11 |  Hy4 preview | Tencent | 55.4 |
| 12 | Claude Opus 4.7 (Adaptive) | Anthropic | 54.7 |
| 12 |  GLM-5.2 | Z.AI | 54.7 |
| 14 | Claude Opus 4.6 | Anthropic | 53 |
| 15 | DSdots3-note Preview | Dots Studio | 52.6 |
| 16 | GLM-5.1 | Z.AI | 52.3 |
| 17 | GPT-5.5 | OpenAI | 52.2 |
| 18 | GPT-5.4 | OpenAI | 52.1 |
| 19 | GLM-5 | Z.AI | 50.4 |
| 19 | Muse Spark | Meta | 50.4 |
| 21 | Claude Sonnet 4.6 | Anthropic | 49 |
| 22 |  MiMo-V2.5-Pro | Xiaomi | 48 |
| 23 | TMInkling-Small | Thinking Machines Lab | 47.8 |
| 24 |  Agents-A1 | InternScience | 47.6 |
| 25 | TMInkling | Thinking Machines Lab | 46 |
| 26 | OAOrnith-1.5-397B | Ornith AI | 44.6 |
| 27 |  Qwen3.8 Max | Alibaba | 43.6 |
| 28 |  DeepSeek V4 Pro 0813 | DeepSeek | 42.7 |
| 29 | GPT-5.4 mini | OpenAI | 41.5 |
| 30 | Qwen3.7 Max | Alibaba | 41.4 |
| 31 |  Gemini 3.5 Flash | 40.2 | |
| 32 | GPT-5.4 nano | OpenAI | 37.7 |
| 33 | DeepSeek V4.1 Flash | DeepSeek | 36.8 |
| 34 | Qwen3.8-Omni-Flash | Alibaba | 36.5 |
| 35 | Qwen3.8-Flash-Next | Alibaba | 35.9 |
| 36 |  Grok 4.3 | xAI | 35 |
| 37 | DeepSeek V4 Flash 0731 | DeepSeek | 34.8 |
| 38 | Kimi K2.6 | Moonshot AI | 34.7 |
| 38 | Qwen3.7 Plus | Alibaba | 34.7 |
| 40 | Claude Opus 4.5 | Anthropic | 30.8 |
| 40 | Qwen3.8-27B | Alibaba | 30.8 |
| 42 | Kimi K2.5 | Moonshot AI | 30.1 |
| 43 | Qwen3.6 Plus | Alibaba | 28.8 |
| 44 | Qwen3.5 397B | Alibaba | 28.7 |
| 45 | STA.X K2 | SK Telecom | 27.8 |
| 46 |  Nemotron 3 Ultra | NVIDIA | 26.7 |
| 47 | Gemma 4 31B | 26.5 | |
| 48 | OAOrnith-1.5-35B-A3B | Ornith AI | 25.6 |
| 49 | Hy3 Preview | Tencent | 25.5 |
| 50 | GLM-4.7 | Z.AI | 24.8 |
| 51 | Qwen3.6-27B | Alibaba | 24 |
| 52 |  Ling 3.0 Flash | InclusionAI | 22.7 |
| 53 | Qwen3.6-35B-A3B | Alibaba | 21.4 |
| 54 | OAOrnith-1.5-9B | Ornith AI | 20.2 |
| 55 | Gemini 2.5 Pro | 18.8 | |
| 56 |  K-EXAONE 2.0 | LG AI Research | 18.3 |
| 57 | Gemma 4 26B A4B | 17.2 | |
| 58 | OPMiniCPM5-2B | OpenBMB | 8.9 |
Evidence key: Observed
Rows are ordered by the value Humanity's Last Exam published, highest first. The source does not state whether a higher value is the better result, so this page does not either: for a benchmark that measures a rate of failure, read the table from the bottom.