Featherbench - Rubric Quality: leaderboard
The judged track of Featherbench — an LLM judge rates answer quality 0-10 on correctness and helpfulness, separating how well a task was done from whether it passed.
Metric: Rubric Score (0-10). Source: reinvently.co.uk. Saturation forecast: Not forecast. 34 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Kimi K3 | 9.5 | #36 |
| 2 | Claude Opus 5.5 | 9.5 | #3 |
| 3 | Claude Fable 5.1 | 9.5 | #10 |
| 4 | Claude Opus 5 | 9.4 | #15 |
| 5 | Claude Fable 5 | 9.3 | #19 |
| 6 | GLM-5.3 | 9.3 | #45 |
| 7 | GLM-5.3 Flash | 9.2 | #79 |
| 8 | Gemini 4 Argon | 9.2 | #9 |
| 9 | Claude Sonnet 5.5 | 9.2 | #13 |
| 10 | Claude Haiku 5.5 | 9.2 | #46 |
| 11 | Beam 1.0 | 9 | #107 |
| 12 | Muse Spark Open | 9 | #41 |
| 13 | Muse Wonder | 9 | #16 |
| 14 | Claude Sonnet 4.6 | 8.9 | #97 |
| 15 | Gemini 3.8 Flash | 8.9 | #30 |
Interactive version: theaggregate.ai/benchmark?slug=featherbench-rubric-quality · How It Works · Data refreshed daily, snapshot 2026-11-07.