MCJudgeBench - Intrinsic Inconsistency: leaderboard
Metric: Intrinsic constraint-level inconsistency rate (%, CIR intr): share of constraints whose predicted label is not identical across 5 repeated judgments of the same instance at temperature 1.0; the judge sees an instruction, a candidate response and its explicit constraint list and labels each constraint yes, partial or no, over the 653 constraints of 141 multi-constraint instructions of MCJudgeBench; lower is better. Source: arxiv.org. Saturation forecast: Estimated already saturated. 11 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Haiku 4.5 | 2.94 |
| 2 | Claude Sonnet 4.6 (Thinking) | 5.21 |
| 3 | Claude Sonnet 4.6 | 5.72 |
| 4 | Gemini 3.1 Pro (Preview) | 6.58 |
| 5 | Gemini 2.5 Flash Lite | 7.35 |
| 6 | GPT-5.2 (Non-reasoning) | 9.49 |
| 7 | Claude Haiku 4.5 (Thinking) | 14.09 |
| 8 | GPT-5.2 | 14.7 |
| 9 | Gemini 2.5 Flash Lite (Thinking) | 14.77 |
| 10 | Qwen 3.5 4B (Non-reasoning) | 24.2 |
| 11 | Llama 3.2 3B Instruct | 35.38 |
Interactive version: theaggregate.ai/benchmark?slug=mcjudgebench-intrinsic-inconsistency · How It Works · Data refreshed daily, snapshot 2026-10-07.