Why this benchmark is useful
When you need a hard, cross-domain expert exam — math through professional law and medicine — as a closed-book stress test, not a research-productivity proxy.
Humanity's Last Exam
HLE
Cross-domain expert-level questions — math, sciences, humanities, professional law and medicine — designed to be the hardest public exam a frontier model can take.
What this does not measure
- Real research productivity — exam-style questions do not capture the messy search-and-synthesize work a researcher actually does.
- Tool-augmented use — HLE is closed-book by default; models that leverage calculators, code, or web search get a separate tier.
- Calibration — HLE reports accuracy, not how well the model knows what it doesn't know.
Coverage map
- Task family
- Reasoning
- Format
- Multiple-choice and short-answer questions vetted by domain experts; submissions are reviewed for difficulty before scoring.
- Scoring
- Percent of questions answered correctly on the public + private split; HLE reports a single accuracy figure per model.
- Maintainer
- Center for AI Safety / Scale AI
How to read the scores
Single accuracy figure on the public + private split. VerdictPal figures usually come from Artificial Analysis' independent run; tool-augmented tiers are separate. Accuracy is not calibration.
What it does not cover
- Real research productivity — exam-style questions do not capture the messy search-and-synthesize work a researcher actually does.
- Tool-augmented use — HLE is closed-book by default; models that leverage calculators, code, or web search get a separate tier.
- Calibration — HLE reports accuracy, not how well the model knows what it doesn't know.
Scores
180 rowsEvidence ledger
180180 rows
55.5%best score
180source-checked
1sources
2026-09-01source date
Where the rows land
0255075100
Normalized to this benchmark's axis (0–100). Open the table for raw units.
Newest receipts
- Claude Fable 555.5% ·
- Claude Opus 5 (Adaptive Reasoning, Max Effort)54.9% ·
- Claude Opus 5 (Adaptive Reasoning, Xhigh Effort)54.4% ·
- Claude Opus 5 (Adaptive Reasoning, High Effort)52.8% ·
- Claude Opus 5 (Adaptive Reasoning, Medium Effort)51.3% ·
Scores use this benchmark's own unit and axis, not a universal quality score.
| # | Model | Release date | Score | Provenance | Trust |
|---|---|---|---|---|---|
| 1 | Claude Fable 5Anthropic | 2026-06-09 | 55.5% | 2026-09-01 | Source-checked |
| 2 | Claude Opus 5 (Adaptive Reasoning, Max Effort)Anthropic | 2026-07-24 | 54.9% | 2026-09-01 | Source-checked |
| 3 | Claude Opus 5 (Adaptive Reasoning, Xhigh Effort)Anthropic | 2026-07-24 | 54.4% | 2026-09-01 | Source-checked |
| 4 | Claude Opus 5 (Adaptive Reasoning, High Effort)Anthropic | 2026-07-24 | 52.8% | 2026-09-01 | Source-checked |
| 5 | Claude Opus 5 (Adaptive Reasoning, Medium Effort)Anthropic | 2026-07-24 | 51.3% | 2026-09-01 | Source-checked |
| 6 | GPT-5.6 SolOpenAI | 2026-07-09 | 49.5% | 2026-09-01 | Source-checked |
| 7 | Claude Opus 4.8Anthropic | 2026-05-28 | 48.7% | 2026-09-01 | Source-checked |
| 8 | Gemini 3.7 FlashGoogle | 2026-08-13 | 47.9% | 2026-09-01 | Source-checked |
| 9 | Gemini 3.1 Pro PreviewGoogle | 2026-02-19 | 47% | 2026-09-01 | Source-checked |
| 10 | Kimi K3Moonshot | 2026-07-16 | 46.9% | 2026-09-01 | Source-checked |
| 11 | Muse Spark 1.1Meta | 2026-07-09 | 46.2% | 2026-09-01 | Source-checked |
| 12 | GPT 5.5OpenAI | 2026-04-23 | 45.8% | 2026-09-01 | Source-checked |
| 13 | Muse Spark 1.2Meta | 2026-08-05 | 45.5% | 2026-09-01 | Source-checked |
| 14 | GPT-5.5 (high)OpenAI | 2026-04-23 | 45% | 2026-09-01 | Source-checked |
| 15 | GPT 5.4OpenAI | 2026-03-05 | 43.7% | 2026-09-01 | Source-checked |
| 16 | Claude Opus 5 (Adaptive Reasoning, Low Effort)Anthropic | 2026-07-24 | 43.4% | 2026-09-01 | Source-checked |
| 17 | Qwen3.8 MaxAlibaba | 2026-08-03 | 43% | 2026-09-01 | Source-checked |
| 18 | GPT-5.6 TerraOpenAI | 2026-07-09 | 42.9% | 2026-09-01 | Source-checked |
| 19 | Grok 4.6xAI | 2026-08-12 | 42.9% | 2026-09-01 | Source-checked |
| 20 | Gemini 3.5 FlashGoogle | 2026-05-19 | 42.7% | 2026-09-01 | Source-checked |
| 21 | Grok 4.5xAI | 2026-07-08 | 42.7% | 2026-09-01 | Source-checked |
| 22 | GPT-5.3 Codex (xhigh)OpenAI | 2026-02-05 | 42.5% | 2026-09-01 | Source-checked |
| 23 | GPT-5.5 (medium)OpenAI | 2026-04-23 | 42.4% | 2026-09-01 | Source-checked |
| 24 | Qwen3.8 2.4T A95BAlibaba | 2026-08-12 | 42.4% | 2026-09-01 | Source-checked |
| 25 | Claude Opus 4.7Anthropic | 2026-04-16 | 42.3% | 2026-09-01 | Source-checked |
| 26 | GLM-5.3Zhipu | 2026-08-14 | 42.3% | 2026-09-01 | Source-checked |
| 27 | Claude Sonnet 5Anthropic | 2026-06-30 | 41.3% | 2026-09-01 | Source-checked |
| 28 | Gemini 3.5 Flash (medium)Google | 2026-05-19 | 41.3% | 2026-09-01 | Source-checked |
| 29 | GLM-5.2Zhipu | 2026-06-13 | 41.1% | 2026-09-01 | Source-checked |
| 30 | DeepSeek V4 Pro 0813 (Reasoning, Max Effort)DeepSeek | 2026-08-13 | 41% | 2026-09-01 | Source-checked |
| 31 | Gemini 3.6 Flash (high)Google | 2026-07-21 | 40.8% | 2026-09-01 | Source-checked |
| 32 | Muse SparkOther | 2026-01-01 | 40.7% | 2026-09-01 | Source-checked |
| 33 | Qwen3.7 MaxAlibaba | 2026-05-19 | 40.5% | 2026-09-01 | Source-checked |
| 34 | Motif 3 (Beta)Other | 2026-07-21 | 40.4% | 2026-09-01 | Source-checked |
| 35 | Claude Opus 4.6 (Adaptive Reasoning, Max Effort)Anthropic | 2026-02-05 | 39.9% | 2026-09-01 | Source-checked |
| 36 | GLM-5.3-FlashZhipu | 2026-08-26 | 39.9% | 2026-09-01 | Source-checked |
| 37 | Gemini 3 ProGoogle | 2025-11-18 | 39.7% | 2026-09-01 | Source-checked |
| 38 | GPT-5.6 LunaOpenAI | 2026-07-09 | 39.5% | 2026-09-01 | Source-checked |
| 39 | MiniMax M3MiniMax | 2026-05-31 | 39% | 2026-09-01 | Source-checked |
| 40 | DeepSeek V4 Flash 0731 (Reasoning, Max Effort)DeepSeek | 2026-07-31 | 38.6% | 2026-09-01 | Source-checked |
| 41 | Grok Build 0.1 0616xAI | 2026-06-16 | 38.3% | 2026-09-01 | Source-checked |
| 42 | Qwen3.8-Flash-NextAlibaba | 2026-08-26 | 38% | 2026-09-01 | Source-checked |
| 43 | GPT-5.2OpenAI | 2025-12-11 | 37.7% | 2026-09-01 | Source-checked |
| 44 | Agnes 2.5 Pro BetaOther | 2026-08-26 | 37.5% | 2026-09-01 | Source-checked |
| 45 | Kimi K2.6Moonshot | 2026-04-20 | 37.5% | 2026-09-01 | Source-checked |
| 46 | Grok 4.3xAI | 2026-04-30 | 37.2% | 2026-09-01 | Source-checked |
| 47 | Motif 3Other | 2026-08-12 | 37% | 2026-09-01 | Source-checked |
| 48 | Gemini 3 Flash Preview (Reasoning)Google | 2025-12-17 | 36.6% | 2026-09-01 | Source-checked |
| 49 | GPT-5.2 Codex (xhigh)OpenAI | 2025-12-11 | 35.7% | 2026-09-01 | Source-checked |
| 50 | MiMo-V2.5-ProXiaomi | 2026-04-22 | 35.7% | 2026-09-01 | Source-checked |
| 51 | Qwen 3.7 PlusAlibaba | 2026-06-02 | 35.6% | 2026-09-01 | Source-checked |
| 52 | DeepSeek V4 Pro (Reasoning, High Effort)DeepSeek | 2026-04-24 | 35.2% | 2026-09-01 | Source-checked |
| 53 | Kimi K2.7 CodeMoonshot | 2026-06-12 | 35% | 2026-09-01 | Source-checked |
| 54 | DeepSeek V4 Flash Vision (Reasoning, Max Effort)DeepSeek | 2026-08-21 | 34.5% | 2026-09-01 | Source-checked |
| 55 | Grok 4.20 ReasoningxAI | 2026-03-05 | 34.5% | 2026-09-01 | Source-checked |
| 56 | Apodex 1.1Other | 2026-08-30 | 34.1% | 2026-09-01 | Source-checked |
| 57 | Qwen3.8 27BAlibaba | 2026-08-14 | 33.9% | 2026-09-01 | Source-checked |
| 58 | LongCat-2.0Meituan | 2026-06-29 | 33.7% | 2026-09-01 | Source-checked |
| 59 | Nex-N2-ProOther | 2026-06-02 | 33.7% | 2026-09-01 | Source-checked |
| 60 | Agnes 2.5 Pro AlphaOther | 2026-07-24 | 33.6% | 2026-09-01 | Source-checked |
| 61 | Claude Sonnet 4.6 (Adaptive Reasoning, Max Effort)Anthropic | 2026-02-17 | 33.6% | 2026-09-01 | Source-checked |
| 62 | Hy3Other | 2026-07-06 | 33.5% | 2026-09-01 | Source-checked |
| 63 | Claude Opus 4.7 (Non-reasoning, High Effort)Anthropic | 2026-04-16 | 33.3% | 2026-09-01 | Source-checked |
| 64 | Inkling SmallOther | 2026-07-30 | 33.3% | 2026-09-01 | Source-checked |
| 65 | GPT-5.5 (low)OpenAI | 2026-04-23 | 32.7% | 2026-09-01 | Source-checked |
| 66 | Grok 4.20 0309 (Reasoning)xAI | 2026-03-10 | 32.4% | 2026-09-01 | Source-checked |
| 67 | Inkling (xhigh)Other | 2026-07-15 | 31.9% | 2026-09-01 | Source-checked |
| 68 | GPT-5.4 (low)OpenAI | 2026-03-05 | 30.8% | 2026-09-01 | Source-checked |
| 69 | Qwen3.6 Max PreviewAlibaba | 2026-04-20 | 30.8% | 2026-09-01 | Source-checked |
| 70 | Kimi K2.5Moonshot | 2026-01-27 | 30.7% | 2026-09-01 | Source-checked |
| 71 | MiMo-V2-ProXiaomi | 2026-03-18 | 30.4% | 2026-09-01 | Source-checked |
| 72 | DeepSeek V4 Flash (Reasoning, High Effort)DeepSeek | 2026-04-24 | 30.3% | 2026-09-01 | Source-checked |
| 73 | Claude Opus 4.5 (Reasoning)Anthropic | 2025-11-24 | 30.1% | 2026-09-01 | Source-checked |
| 74 | GLM-5.1Zhipu | 2026-04-07 | 30.1% | 2026-09-01 | Source-checked |
| 75 | Grok 4.3 (medium)xAI | 2026-04-30 | 30% | 2026-09-01 | Source-checked |
| 76 | A.X-K2Other | 2026-08-12 | 29.6% | 2026-09-01 | Source-checked |
| 77 | MiniMax M2.7MiniMax | 2026-04-15 | 29.6% | 2026-09-01 | Source-checked |
| 78 | Gemini 3 Pro Preview (low)Google | 2025-11-18 | 29.5% | 2026-09-01 | Source-checked |
| 79 | GLM-5 (Reasoning)Zhipu | 2026-02-11 | 29.3% | 2026-09-01 | Source-checked |
| 80 | Solar Pro 4Other | 2026-08-06 | 29.2% | 2026-09-01 | Source-checked |
| 81 | Qwen3.5 397B A17B (Reasoning)Alibaba | 2026-02-16 | 29% | 2026-09-01 | Source-checked |
| 82 | GPT-5 (high)OpenAI | 2025-08-07 | 28.5% | 2026-09-01 | Source-checked |
| 83 | GPT-5.1OpenAI | 2025-11-13 | 28.5% | 2026-09-01 | Source-checked |
| 84 | Solar Open2 250BOther | 2026-08-12 | 28.5% | 2026-09-01 | Source-checked |
| 85 | Nemotron 3 Ultra 550B A55B (Reasoning)NVIDIA | 2026-06-04 | 28.4% | 2026-09-01 | Source-checked |
| 86 | GPT-5.4 NanoOpenAI | 2026-03-17 | 28.3% | 2026-09-01 | Source-checked |
| 87 | GPT-5.4 MiniOpenAI | 2026-03-17 | 28.1% | 2026-09-01 | Source-checked |
| 88 | Qwen3 Max ThinkingAlibaba | 2026-01-26 | 28% | 2026-09-01 | Source-checked |
| 89 | Grok 4.20 0309 v2 (Non-reasoning)xAI | 2026-04-07 | 27.9% | 2026-09-01 | Source-checked |
| 90 | GLM-5-TurboZhipu | 2026-03-15 | 27.8% | 2026-09-01 | Source-checked |
| 91 | GPT-5 Codex (high)OpenAI | 2025-09-23 | 27.8% | 2026-09-01 | Source-checked |
| 92 | Hy3-preview (Reasoning)Other | 2026-04-23 | 27.8% | 2026-09-01 | Source-checked |
| 93 | Qwen3.6 PlusAlibaba | 2026-04-02 | 27.8% | 2026-09-01 | Source-checked |
| 94 | GLM-4.7 (Reasoning)Zhipu | 2025-12-22 | 27.4% | 2026-09-01 | Source-checked |
| 95 | MiMo-V2.5Xiaomi | 2026-04-22 | 27.2% | 2026-09-01 | Source-checked |
| 96 | GPT-5.2 (medium)OpenAI | 2025-12-11 | 26.7% | 2026-09-01 | Source-checked |
| 97 | Grok 4xAI | 2025-07-10 | 26.7% | 2026-09-01 | Source-checked |
| 98 | GPT-5.1 Codex (high)OpenAI | 2025-11-13 | 25.7% | 2026-09-01 | Source-checked |
| 99 | GPT-5 (medium)OpenAI | 2025-08-07 | 25.4% | 2026-09-01 | Source-checked |
| 100 | Qwen3.5 122B A10B (Reasoning)Alibaba | 2026-02-24 | 25.2% | 2026-09-01 | Source-checked |
| 101 | Kimi K3 (low)Moonshot | 2026-07-16 | 25% | 2026-09-01 | Source-checked |
| 102 | DeepSeek V3.2 (Reasoning)DeepSeek | 2025-12-01 | 24.6% | 2026-09-01 | Source-checked |
| 103 | Gemini 3.5 Flash (minimal)Google | 2026-05-19 | 24.1% | 2026-09-01 | Source-checked |
| 104 | Qwen3.5 27B (Reasoning)Alibaba | 2026-02-24 | 23.9% | 2026-09-01 | Source-checked |
| 105 | Kimi K2 ThinkingMoonshot | 2025-11-06 | 23.8% | 2026-09-01 | Source-checked |
| 106 | Ling 3.0 FlashOther | 2026-08-04 | 23.7% | 2026-09-01 | Source-checked |
| 107 | Gemma 4 31B ITGoogle | 2026-03-01 | 23.6% | 2026-09-01 | Source-checked |
| 108 | MiniMax-M2.1MiniMax | 2025-12-23 | 23.2% | 2026-09-01 | Source-checked |
| 109 | Qwen3.6 27B (Reasoning)Alibaba | 2026-04-22 | 23.1% | 2026-09-01 | Source-checked |
| 110 | MiMo-V2-Flash (Reasoning)Xiaomi | 2025-12-16 | 22.8% | 2026-09-01 | Source-checked |
| 111 | MiMo-V2-Omni-0327Xiaomi | 2026-03-27 | 22.6% | 2026-09-01 | Source-checked |
| 112 | Qwen3.6 35B A3B (Reasoning)Alibaba | 2026-04-16 | 22.2% | 2026-09-01 | Source-checked |
| 113 | MiMo-V2-Flash (Feb 2026)Xiaomi | 2025-12-16 | 22.1% | 2026-09-01 | Source-checked |
| 114 | MiMo-V2-OmniXiaomi | 2026-03-19 | 22.1% | 2026-09-01 | Source-checked |
| 115 | Muse GlimmerMeta | 2026-08-10 | 22% | 2026-09-01 | Source-checked |
| 116 | GPT-5.5 Instant (May 2026)OpenAI | 2026-05-05 | 21.6% | 2026-09-01 | Source-checked |
| 117 | Ring-2.6-1TOther | 2026-05-08 | 21.6% | 2026-09-01 | Source-checked |
| 118 | Step 3.7 FlashStepFun | 2026-06-01 | 21.4% | 2026-09-01 | Source-checked |
| 119 | MiniMax M2.5MiniMax | 2026-04-01 | 20.5% | 2026-09-01 | Source-checked |
| 120 | OpenAI o3OpenAI | 2025-04-16 | 20.1% | 2026-09-01 | Source-checked |
| 121 | GPT-5.5 Instant (June 2026)OpenAI | 2026-06-25 | 19.9% | 2026-09-01 | Source-checked |
| 122 | Qwen3.5 397B A17B (Non-reasoning)Alibaba | 2026-02-16 | 19.8% | 2026-09-01 | Source-checked |
| 123 | GPT-5 (low)OpenAI | 2025-08-07 | 19.6% | 2026-09-01 | Source-checked |
| 124 | Kimi K2.6 (Non-reasoning)Other | 2026-04-20 | 19.6% | 2026-09-01 | Source-checked |
| 125 | Claude Opus 4.6Anthropic | 2026-02-05 | 19.1% | 2026-09-01 | Source-checked |
| 126 | Gemini 3.5 Flash-LiteGoogle | 2026-07-21 | 18.8% | 2026-09-01 | Source-checked |
| 127 | K-EXAONE 2.0 0803Other | 2026-08-12 | 18.6% | 2026-09-01 | Source-checked |
| 128 | Grok 4.3 (low)xAI | 2026-04-30 | 18.4% | 2026-09-01 | Source-checked |
| 129 | Claude 4.5 Sonnet (Reasoning)Anthropic | 2025-09-29 | 17.8% | 2026-09-01 | Source-checked |
| 130 | JT-4.1 Flash 236B A21BOther | 2026-07-09 | 17.8% | 2026-09-01 | Source-checked |
| 131 | G9v3-39A5BOther | 2026-08-03 | 17.5% | 2026-09-01 | Source-checked |
| 132 | Gemini 3.1 Flash LiteGoogle | 2026-05-07 | 17.2% | 2026-09-01 | Source-checked |
| 133 | GLM 5V Turbo (Reasoning)Zhipu | 2026-04-01 | 17.1% | 2026-09-01 | Source-checked |
| 134 | HyperNova 60B 2605Multiverse | 2026-05-06 | 16.6% | 2026-09-01 | Source-checked |
| 135 | KAT Coder Pro V2Other | 2026-03-27 | 16.1% | 2026-09-01 | Source-checked |
| 136 | DeepSeek R1DeepSeek | 2025-01-20 | 15.8% | 2026-09-01 | Source-checked |
| 137 | Gemma 4 12B (Reasoning)Google | 2026-06-07 | 15.7% | 2026-09-01 | Source-checked |
| 138 | Qwen3.6 27B (Non-reasoning)Alibaba | 2026-04-22 | 15.1% | 2026-09-01 | Source-checked |
| 139 | Gemini 3 Flash PreviewGoogle | 2025-12-17 | 15% | 2026-09-01 | Source-checked |
| 140 | MiMo-V2.5-Pro (Non-reasoning)Xiaomi | 2026-04-22 | 14.8% | 2026-09-01 | Source-checked |
| 141 | Qwen3.6 35B A3B (Non-reasoning)Alibaba | 2026-04-16 | 13.9% | 2026-09-01 | Source-checked |
| 142 | Mistral Medium 3.5Mistral | 2026-04-29 | 13.8% | 2026-09-01 | Source-checked |
| 143 | GPT-5.5 (Non-reasoning)OpenAI | 2026-04-23 | 13.7% | 2026-09-01 | Source-checked |
| 144 | Claude Sonnet 4.6Anthropic | 2026-02-17 | 13.3% | 2026-09-01 | Source-checked |
| 145 | Claude Opus 4.5Anthropic | 2025-11-24 | 13.2% | 2026-09-01 | Source-checked |
| 146 | EXAONE 4.5 33BOther | 2026-04-09 | 12.9% | 2026-09-01 | Source-checked |
| 147 | Claude 4.1 Opus (Reasoning)Anthropic | 2025-08-05 | 12.5% | 2026-09-01 | Source-checked |
| 148 | Claude 4 Opus (Reasoning)Anthropic | 2025-05-22 | 12.3% | 2026-09-01 | Source-checked |
| 149 | Command ACohere | 2026-03-15 | 12% | 2026-09-01 | Source-checked |
| 150 | Claude Sonnet 4.6 (Non-reasoning, Low Effort)Anthropic | 2026-02-17 | 11.2% | 2026-09-01 | Source-checked |
| 151 | Granite 4.2 30BOther | 2026-08-25 | 11.2% | 2026-09-01 | Source-checked |
| 152 | North Mini CodeCohere | 2026-06-09 | 11.1% | 2026-09-01 | Source-checked |
| 153 | DiffusionGemma 26B A4BGoogle | 2026-06-10 | 10.8% | 2026-09-01 | Source-checked |
| 154 | Nemotron 3.5 LightningNVIDIA | 2026-08-11 | 10.6% | 2026-09-01 | Source-checked |
| 155 | GLM-5.2 (Non-reasoning)Zhipu | 2026-06-16 | 9.8% | 2026-09-01 | Source-checked |
| 156 | Granite 4.2 8BOther | 2026-08-25 | 9.7% | 2026-09-01 | Source-checked |
| 157 | Ling 3.0 TinyOther | 2026-08-06 | 9.3% | 2026-09-01 | Source-checked |
| 158 | Ling-2.6-1TOther | 2026-04-23 | 8.7% | 2026-09-01 | Source-checked |
| 159 | DeepSeek V4 Pro (Non-reasoning)DeepSeek | 2026-04-24 | 8.2% | 2026-09-01 | Source-checked |
| 160 | DeepSeek V4 Flash (Non-reasoning)DeepSeek | 2026-04-24 | 7.8% | 2026-09-01 | Source-checked |
| 161 | GLM-5 (Non-reasoning)Zhipu | 2026-02-11 | 7.6% | 2026-09-01 | Source-checked |
| 162 | Hy3-preview (Non-reasoning)Other | 2026-04-23 | 7% | 2026-09-01 | Source-checked |
| 163 | OpenAI o1OpenAI | 2024-09-12 | 7% | 2026-09-01 | Source-checked |
| 164 | LFM2.5-8B-A1BLiquid | 2026-06-08 | 6.9% | 2026-09-01 | Source-checked |
| 165 | Grok 4.3 (Non-reasoning)xAI | 2026-04-30 | 6.8% | 2026-09-01 | Source-checked |
| 166 | Granite 4.2 3BOther | 2026-08-25 | 6.6% | 2026-09-01 | Source-checked |
| 167 | MiniCPM5-1B (Reasoning)OpenBMB | 2026-06-04 | 6.5% | 2026-09-01 | Source-checked |
| 168 | JT-35B-FlashOther | 2026-05-14 | 6.4% | 2026-09-01 | Source-checked |
| 169 | JT-MINIOther | 2026-04-15 | 6.4% | 2026-09-01 | Source-checked |
| 170 | Gemma 4 12B (Non-reasoning)Google | 2026-06-10 | 6.3% | 2026-09-01 | Source-checked |
| 171 | Ling 2.6 FlashOther | 2026-04-21 | 6.3% | 2026-09-01 | Source-checked |
| 172 | MiniCPM-V 4.6 1.3BOpenBMB | 2026-05-11 | 5.1% | 2026-09-01 | Source-checked |
| 173 | Llama 4 MaverickMeta | 2026-04-05 | 4.9% | 2026-09-01 | Source-checked |
| 174 | Nemotron 3 Nano Omni 30B A3B ReasoningNVIDIA | 2026-04-29 | 4.8% | 2026-09-01 | Source-checked |
| 175 | G9v3-3BOther | 2026-07-23 | 4.5% | 2026-09-01 | Source-checked |
| 176 | MiniCPM5-1B (Non-reasoning)OpenBMB | 2026-05-25 | 4.5% | 2026-09-01 | Source-checked |
| 177 | Mistral Large 3Mistral | 2025-12-02 | 4.2% | 2026-09-01 | Source-checked |
| 178 | Granite 4.1 30BOther | 2026-04-29 | 4.1% | 2026-09-01 | Source-checked |
| 179 | Llama 4 ScoutMeta | 2026-04-05 | 3.8% | 2026-09-01 | Source-checked |
| 180 | Claude 3 OpusAnthropic | 2024-03-04 | 2.8% | 2026-09-01 | Source-checked |
What it covers
This benchmark sits in the reasoning family. It uses Multiple-choice and short-answer questions vetted by domain experts; submissions are reviewed for difficulty before scoring. The score should travel with its task format, scoring method, source date, and benchmark version.
Where it breaks down
Scores on VerdictPal come from Artificial Analysis' independent HLE run unless noted. The official project site hosts the dataset and paper — not a single maintained public leaderboard URL.
Receipts
Sources and further reading
- Humanity's Last Exam project· checked 2026-06-09
- Artificial Analysis HLE evaluation· checked 2026-06-09
- HLE paper (arXiv)· checked 2026-06-09
- Project site
- AA evaluation
Questions about this benchmark
Every answer below is assembled from the dated fields on this page. Nothing is written separately for search.
What does Humanity's Last Exam measure?
Cross-domain expert-level questions — math, sciences, humanities, professional law and medicine — designed to be the hardest public exam a frontier model can take.
What does a high Humanity's Last Exam score not prove?
A strong Humanity's Last Exam result says nothing about:
- Real research productivity — exam-style questions do not capture the messy search-and-synthesize work a researcher actually does.
- Tool-augmented use — HLE is closed-book by default; models that leverage calculators, code, or web search get a separate tier.
- Calibration — HLE reports accuracy, not how well the model knows what it doesn't know.
How is Humanity's Last Exam scored?
Percent of questions answered correctly on the public + private split; HLE reports a single accuracy figure per model. Task format: Multiple-choice and short-answer questions vetted by domain experts; submissions are reviewed for difficulty before scoring.
Is Humanity's Last Exam saturated?
Humanity's Last Exam is currently marked Active in the atlas.
Can Humanity's Last Exam results be contaminated by training data?
Contamination risk for Humanity's Last Exam is graded Low contamination. Treat every row on this page as a public claim with a source and a date, not as a controlled experiment.
The Pack · Editorial newsletter
New tools in your inbox. Free.
One short email when a tool ships or changes status. No tracking, no third-party analytics. Unsubscribe in one click.