MM-CondChain: leaderboard
Metric: Path F1 (%), the harmonic mean of True-path accuracy (follow every condition to the final answer) and False-path accuracy (stop at the perturbed condition and pick its auxiliary answer), averaged over the Natural, Chart and GUI domains; zero-shot multiple choice with a boxed answer, unparseable outputs wrong; higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 27 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Pro | 53.33 | #77 |
| 2 | GPT-5 | 50.34 | #91 |
| 3 | Gemini 3 Flash | 48.31 | #93 |
| 4 | Qwen 3 VL 235B A22B (Thinking) | 46.83 | #228 (Qwen 3 VL 235B A22B) |
| 5 | Qwen 3.5 397B A17B | 45.9 | #141 |
| 6 | Qwen 3 VL 235B A22B Instruct | 45.45 | #264 |
| 7 | Kimi K2.5 | 45.25 | #139 |
| 8 | Gemini 2.5 Pro | 40.91 | #145 |
| 9 | Qwen 3 VL 30B A3B (Thinking) | 40.4 | #338 (Qwen 3 VL 30B A3B) |
| 10 | Qwen 3.5 122B A10B | 40.12 | #170 |
| 11 | Qwen 3 VL 8B (Thinking) | 38.25 | |
| 12 | GLM-4.6V | 36.99 | #309 |
| 13 | Qwen 3 VL 8B Instruct | 32.65 | #401 |
| 14 | Gemini 2.5 Flash | 29.64 | #237 |
| 15 | Qwen 3.5 9B | 28.62 | #363 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=mm-condchain · How It Works · Data refreshed daily, snapshot 2026-10-11.