BBH: leaderboard
Big-Bench Hard benchmark with tasks requiring multi-step reasoning.
Metric: Score (imported). Source: neosignal.io. Source: benchmarklist.com. Saturation forecast: Estimated already saturated. 16 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | DeepSeek V3 | 87.5 | #321 |
| 2 | Llama 3.1 405B | 82.9 | #466 |
| 3 | Phi-3-medium-128k-instruct | 81.4 | #1013 |
| 4 | Qwen 2.5 72B | 79.8 | #499 |
| 5 | Phi-3-small-8k-instruct | 79.1 | #967 |
| 6 | GPT-4.1 | 75.12 | #248 |
| 7 | Phi-3 Mini 4K Instruct | 71.7 | #1297 |
| 8 | Llama 2 70B Base | 64.9 | #1025 |
| 9 | GPT-3.5 Turbo (1106) | 61.59 | #961 |
| 10 | Phi-4 | 59.4 | #706 |
| 11 | Llama 2 7B | 58.5 | #1520 |
| 12 | Mistral-7B-v0.1 | 56.1 | #1326 |
| 13 | gemma-7B | 55.1 | #1384 |
| 14 | Qwen 3 235B A22B | 55 | #313 |
| 15 | Yi 6B (Base) | 47.2 | #1472 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=bbh · How It Works · Data refreshed daily, snapshot 2026-10-20.