MCJudgeBench - Response-Variant Inconsistency: leaderboard
Metric: Procedural inconsistency rate under candidate-response variants with parse penalty (%, CIR dagger resp): share of constraint predictions that differ from the reference prediction on 201 human-validated label-preserving response perturbations (local paraphrase, structural reordering), unparsable outputs counted as inconsistent, temperature 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 | Qwen 3.5 4B (Non-reasoning) | 3.59 |
| 2 | Gemini 2.5 Flash Lite | 3.81 |
| 3 | Gemini 3.1 Pro (Preview) | 4.02 |
| 4 | Claude Sonnet 4.6 (Thinking) | 4.02 |
| 5 | Claude Sonnet 4.6 | 4.97 |
| 6 | Claude Haiku 4.5 | 5.29 |
| 7 | Llama 3.2 3B Instruct | 6.17 |
| 8 | Claude Haiku 4.5 (Thinking) | 7.19 |
| 9 | GPT-5.2 | 7.29 |
| 10 | GPT-5.2 (Non-reasoning) | 7.51 |
| 11 | Gemini 2.5 Flash Lite (Thinking) | 8.2 |
Interactive version: theaggregate.ai/benchmark?slug=mcjudgebench-response-variant-inconsistency · How It Works · Data refreshed daily, snapshot 2026-10-07.