MMPCBench - Premise Contradiction PCQ: leaderboard
MMPCBench (arXiv 2608.29286v1, Table 2): proactive critique of flawed multimodal inputs for 14 MLLMs; a judge panel of GPT-5.2, Gemini-3-Pro and Claude-Sonnet-4 scores the responses, and two of those judges are also evaluated. PCQ on inputs with cross-modal, inter-image or intra-text contradictions in their premises; 0-100, higher is better.
Metric: PCQ (0-100). Source: arxiv.org. Saturation forecast: Not forecast. 14 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3 VL 235B A22B Instruct | 51.3 | #280 |
| 2 | Qwen 3 VL 30B A3B Instruct | 40.5 | #390 |
| 3 | Gemini 3 Pro | 39.2 | #82 |
| 4 | Qwen 3 VL 8B Instruct | 37.9 | #419 |
| 5 | Gemini 3 Flash | 35.7 | #103 |
| 6 | Claude Sonnet 4.5 (Thinking) | 17.9 | #148 (Claude Sonnet 4.5) |
| 7 | Gemma 3 27B (IT) | 11.9 | #528 |
| 8 | Gemma 3 12B (IT) | 11.1 | #672 |
| 9 | GPT-5.2 | 10.7 | #115 |
| 10 | GPT-5.1 (High) | 7.2 | #145 (GPT-5.1) |
| 11 | GLM-4.6V | 6.9 | #330 |
| 12 | GPT-5 Mini | 6.6 | #193 |
| 13 | Gemma 3 4B (IT) | 6 | #997 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=mmpcbench-premise-contradiction-pcq · How It Works · Data refreshed daily, snapshot 2026-11-05.