Facet-Probe - Mixed-Modality Order: leaderboard
Metric: Flip rate (% of items whose normalized answer changes across six sampled orderings of text and image evidence in free-form multimodal RAG (MRAMG-Recipe, MMDocRAG, MultiModalQA; LLM-judge correctness); lower is better). Source: arxiv.org. Saturation forecast: Around December 2026. 18 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-5.5 (Medium) | 24 |
| 2 | Gemini 3.1 Pro (Preview) | 25 |
| 3 | Gemini 3 Flash (Preview) | 26 |
| 4 | Claude Opus 4.7 (Thinking) | 26 |
| 5 | Claude Sonnet 4.6 (Thinking) | 30 |
| 6 | GPT-5.4 Mini (Medium) | 36 |
| 7 | Qwen 3.5 9B | 52 |
| 8 | Qwen 3.5 4B | 53 |
| 9 | Qwen 3.5 2B | 71 |
| 10 | MedGemma-4B-IT | 77 |
| 11 | Qwen 3.5 0.8B | 84 |
Interactive version: theaggregate.ai/benchmark?slug=facet-probe-mixed-modality-order · How It Works · Data refreshed daily, snapshot 2026-09-26.