BiGGen-Bench: leaderboard
BiGGen-Bench — fine-grained generation benchmark (77 tasks, 765 instances) scored per-instance by evaluator LMs against instance-specific rubrics.
Metric: Average Score (1-5). Source: huggingface.co. Saturation forecast: Estimated already saturated. 103 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4 Preview (1106) | 4.22 | #418 |
| 2 | GPT-4 Preview (0125) | 4.19 | #486 |
| 3 | GPT-4o (2024-05-13) | 4.14 | #325 |
| 4 | GPT-4 Turbo | 4.13 | #404 |
| 5 | Claude 3 Opus (20240229) | 4.1 | #412 |
| 6 | Llama 3 70B Instruct | 4.01 | #634 |
| 7 | Claude 3 Sonnet (20240229) | 4.01 | #677 |
| 8 | Claude 3 Haiku (20240307) | 3.95 | #771 |
| 9 | Gemini 1.5 Pro | 3.95 | #445 |
| 10 | Mixtral 8x22B Instruct (v0.1) | 3.94 | #756 |
| 11 | Mistral Medium | 3.94 | #792 |
| 12 | Mistral Large | 3.93 | #710 |
| 13 | Gemini 1.5 Flash | 3.9 | #618 |
| 14 | Qwen 1.5 72B Chat | 3.83 | #776 |
| 15 | Phi-3 Mini 4K Instruct | 3.82 | #1297 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=biggen-bench · How It Works · Data refreshed daily, snapshot 2026-10-20.