RubricEval: leaderboard
Instruction-following evaluation framework for open-ended tasks using example-specific rubrics and scalable model grading.
Metric: Mean Score (1-5). Source: huggingface.co. Saturation forecast: Estimated already saturated. 13 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4 | 3.18 | #438 |
| 2 | GPT-4 Turbo | 3.1 | #404 |
| 3 | Gemini 1.5 Pro | 3.06 | #445 |
| 4 | Gemini 1.5 Flash | 2.98 | #618 |
| 5 | Llama 3 70B | 2.9 | #631 |
| 6 | Claude 3 Opus | 2.86 | #571 |
| 7 | Claude 3 Sonnet | 2.79 | #775 |
| 8 | Claude 3 Haiku | 2.73 | #803 |
| 9 | Llama 3 8B | 2.56 | #1242 |
| 10 | Gemini 1.0 Pro | 2.56 | #933 |
| 11 | GPT-3.5 Turbo | 2.52 | #865 |
| 12 | gemma-7B | 2.14 | #1384 |
| 13 | gemma-2B | 1.83 | #1575 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=rubriceval · How It Works · Data refreshed daily, snapshot 2026-10-20.