BlueBench - Chatbot Abilities: leaderboard
BlueBench chatbot task using Arena-Hard-style instruction-following prompts judged against a strong reference response.
Metric: Score (%). Source: huggingface.co. Saturation forecast: Estimated already saturated. 18 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4.1 Mini | 97.55 | #355 |
| 2 | GPT-4.1 | 96.36 | #248 |
| 3 | GPT-4.1 Nano | 95.57 | #731 |
| 4 | Mistral Medium 3 | 94.17 | #500 |
| 5 | GPT-4o | 94.14 | #343 |
| 6 | Mistral Large | 86.13 | #710 |
| 7 | Pixtral-12B | 76.54 | #808 |
| 8 | Granite 3.3 8B Instruct | 60.78 | #825 |
| 9 | Llama 3.3 70B Instruct | 49.69 | #534 |
| 10 | O3 Mini | 37.11 | #272 |
| 11 | O1 | 34.38 | #183 |
| 12 | Llama 3.2 3B Instruct | 25.14 | #1352 |
| 13 | O4 Mini | 11.39 | #180 |
| 14 | Llama 3.2 1B Instruct | 5.7 | #1539 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=bluebench-chatbot-abilities · How It Works · Data refreshed daily, snapshot 2026-10-20.