Big-Bench Hard: leaderboard
23 tasks from BIG-Bench where prior language models failed to outperform average human raters. Tests algorithmic reasoning, language understanding, and world knowledge.
Metric: Average (%). Source: epoch.ai. Saturation forecast: Estimated already saturated. 50 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 1.5 Pro (001) | 89.2 | #444 |
| 2 | DeepSeek V3 | 87.5 | #321 |
| 3 | Llama 3.1 405B | 82.9 | #466 |
| 4 | Phi-3-medium-128k-instruct | 81.4 | #1013 |
| 5 | Qwen 2.5 72B | 79.8 | #499 |
| 6 | Phi-3-small-8k-instruct | 79.1 | #967 |
| 7 | DeepSeek V2 | 78.8 | #549 |
| 8 | GPT-4 (0613) | 75.12 | #491 |
| 9 | Yi Chat 34B | 71.7 | #976 |
| 10 | Phi-3 Mini 4K Instruct | 71.7 | #1297 |
| 11 | StableBeluga2 | 69.3 | #1078 |
| 12 | Llama 2 70B Base | 64.9 | #1025 |
| 13 | GPT-3.5 Turbo (0613) | 61.59 | #1118 |
| 14 | Phi-2 | 59.4 | #1511 |
| 15 | Llama 2 70B Chat | 58.5 | #1237 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=big-bench-hard · How It Works · Data refreshed daily, snapshot 2026-10-20.