Judgemark v2.1 — leaderboard
Evaluates LLM-as-judge quality: stability (self-consistency), separability (ability to distinguish model quality), and correlation with human preferences.
Metric: Judgemark Score (0-100). Source: eqbench.com. Status: saturation imminent. 47 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Sonnet 4.6 | 94.08 |
| 2 | Claude Opus 4.6 | 92.93 |
| 3 | Gemini 3 Pro (Preview) | 86.58 |
| 4 | GLM-5 | 85.58 |
| 5 | Qwen 3.5 397B A17B | 85.45 |
| 6 | Gemini 3 Flash (Preview) | 85.38 |
| 7 | GPT-5.4 | 84.98 |
| 8 | Claude Sonnet 4.5 | 84.56 |
| 9 | Qwen 3.5 27B | 84.31 |
| 10 | GPT-5.2 | 82.01 |
| 11 | Claude Sonnet 4 | 81.99 |
| 12 | Qwen 3.5 35B A3B | 81.39 |
| 13 | GPT-5.4 Mini | 81.14 |
| 14 | Qwen 3.5 122B A10B | 81.03 |
| 15 | Gemini 3.1 Flash Lite (Preview) | 80.92 |
Interactive version: theaggregate.ai/benchmark?slug=judgemark-v2-1 · How the rankings work · Data refreshed daily, snapshot 2026-07-22.