Updated daily at 05:00 UTC
Benchmark
Humanity's Last Exam (HLE)
Measures: 2,500 expert-written, closed-ended questions across 100+ subjects (about 14% need an image); a model gives one final answer plus a confidence, graded for exact-match accuracy by an LLM judge, with calibration error reported alongside.
- Maintainer
- Center for AI Safety (CAIS) and Scale AI
- Introduced
- 2025
- Status
- Active
- Saturation
- Headroom
- Typical provenance
- Mixed
- In scoring set
- Yes
Leaderboard
Every result we hold, whoever ran it. Only independent and third-party rows count towards a score.
| Model | Accuracy | Calibration error | Provenance | Source | Evaluated |
|---|---|---|---|---|---|
| DeepSeek-V4.1-Flash | 63.9% | — | Self-reported | DeepSeek-V4.1-Flash model card via Hugging Face Community Evals | — |
| GLM-5.3 | 62.5% | — | Self-reported | Model Card via Hugging Face Community Evals | — |
| Kimi K3 | 56.0% | — | Self-reported | Model Card via Hugging Face Community Evals | — |
| Hy4 preview | 55.4% | — | Self-reported | Model Card via Hugging Face Community Evals | — |
| GLM-5.3-Flash | 55.3% | — | Self-reported | GLM-5.3-Flash model card via Hugging Face Community Evals | — |
| GLM-5.2 | 54.7% | — | Self-reported | Model Card via Hugging Face Community Evals | — |
| Kimi K2.6 | 54.0% | — | Self-reported | Model Card via Hugging Face Community Evals | — |
| Hy3 | 53.2% | — | Self-reported | Model Card via Hugging Face Community Evals | — |
| GLM-5.1 | 52.3% | — | Self-reported | Model Card via Hugging Face Community Evals | — |
| GLM-5 | 50.4% | — | Self-reported | Model Card via Hugging Face Community Evals | — |
| Kimi K2.5 | 50.2% | — | Self-reported | Model Card via Hugging Face Community Evals | — |
| Qwen3.5-27B | 48.5% | — | Self-reported | Model Card via Hugging Face Community Evals | — |
| Qwen3.5-397B-A17B | 48.3% | — | Self-reported | Model Card via Hugging Face Community Evals | — |
| Step 3.7 Flash | 48.1% | — | Self-reported | Model Card via Hugging Face Community Evals | — |
| MiMo-V2.5-Pro | 48.0% | — | Self-reported | Model Card via Hugging Face Community Evals | — |
| Qwen3.5-122B-A10B | 47.5% | — | Self-reported | Model Card via Hugging Face Community Evals | — |
| Inkling | 46.0% | — | Self-reported | Model Card via Hugging Face Community Evals | — |
| Kimi K2 Thinking Moonshot AI · Thinking effort | 44.9% | — | Self-reported | Model Card via Hugging Face Community Evals | — |
| Qwen3.8-2.4T-A95B | 43.6% | — | Self-reported | Model Card via Hugging Face Community Evals | — |
| GLM-4.7 | 42.8% | — | Self-reported | Model Card via Hugging Face Community Evals | — |
| Nemotron 3 Ultra 550B-A55B | 37.4% | — | Self-reported | Model Card via Hugging Face Community Evals | — |
| Qwen3.8-Flash-Next | 35.9% | — | Self-reported | Qwen3.8-Flash-Next model card via Hugging Face Community Evals | — |
| Inkling Small | 31.6% | — | Self-reported | Model Card via Hugging Face Community Evals | — |
| Qwen3.8-27B | 30.8% | — | Self-reported | Qwen3.8-27B model card via Hugging Face Community Evals | — |
| Hy3 preview | 30.0% | — | Self-reported | Model Card via Hugging Face Community Evals | — |
| Solar Open 2 (250B) | 28.8% | — | Self-reported | Model Card via Hugging Face Community Evals | — |
| A.X K2 | 27.8% | — | Self-reported | Model Card via Hugging Face Community Evals | — |
| Qwen3.6-27B | 24.0% | — | Self-reported | Model Card via Hugging Face Community Evals | — |
| Step 3.5 Flash | 23.1% | — | Self-reported | Step 3.5 Flash Paper via Hugging Face Community Evals | — |
| Nemotron 3 Super 120B-A12B | 22.8% | — | Self-reported | Model Card via Hugging Face Community Evals | — |
| Ling-3.0-flash | 22.7% | — | Self-reported | Model Card via Hugging Face Community Evals | — |
| MiniMax-M2.1 | 22.2% | — | Self-reported | Model Card via Hugging Face Community Evals | — |
| MiMo-V2-Flash | 22.1% | — | Self-reported | Model Card via Hugging Face Community Evals | — |
| Muse Glimmer 30B | 22.0% | — | Self-reported | Muse Glimmer-30B model card via Hugging Face Community Evals | — |
| Intern-S2-Preview | 21.9% | — | Self-reported | Model Card via Hugging Face Community Evals | — |
| Qwen3.6-35B-A3B | 21.4% | — | Self-reported | Model Card via Hugging Face Community Evals | — |
| MiniMax-M2.5 | 19.4% | — | Self-reported | Model Card via Hugging Face Community Evals | — |
| gpt-oss-120b | 19.0% | — | Self-reported | GPT-OSS Model Card via Hugging Face Community Evals | — |
| gpt-oss-20b | 17.3% | — | Self-reported | GPT-OSS Model Card via Hugging Face Community Evals | — |
| GLM-4.7-Flash | 14.4% | — | Self-reported | Model Card via Hugging Face Community Evals | — |
| K-EXAONE-236B-A23B | 13.6% | — | Self-reported | Model Card via Hugging Face Community Evals | — |
| MiniMax-M2 | 12.5% | — | Self-reported | Model Card via Hugging Face Community Evals | — |
| Nemotron 3.5 Lightning 30B-A3B | 11.7% | — | Self-reported | NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16 model card via Hugging Face Community Evals | — |
| A.X K1 | 8.6% | — | Self-reported | A.X K1 model card via Hugging Face Community Evals | — |
| GLM-4.5 | 8.3% | — | Self-reported | EvalEval via Hugging Face Community Evals | — |
| GLM-4.5-Air | 8.1% | — | Self-reported | EvalEval via Hugging Face Community Evals | — |
- Reported on Hugging Face Community Evals; each result names its author and links its source.