MCJudgeBench: leaderboard
Multi-constraint instruction following requires verifying whether a response satisfies multiple individual requirements, yet LLM judges are often assessed only through overall-response judgments. We introduce MCJudgeBench, a benchmark for constraint-level judge evaluation in multi-constraint instruction following. Each instance includes an instruction, a candidate response, an explicit constraint
Metric: CJAR (self-reported). Source: benchmarklist.com. Saturation forecast: Around July 2027. 7 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 85.8 | #58 |
| 2 | Gemini 2.5 Flash Lite | 85.5 | #422 |
| 3 | Qwen 3.5 4B | 85.3 | #475 |
| 4 | Claude Haiku 4.5 | 84.8 | #280 |
| 5 | Claude Sonnet 4.6 | 82.8 | #90 |
| 6 | GPT-5.2 | 80.9 | #110 |
| 7 | Llama 3.2 3B Instruct | 77 | #1352 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=mcjudgebench · How It Works · Data refreshed daily, snapshot 2026-10-20.