ProactiveBench - ChangeIt: leaderboard
Metric: Trajectory accuracy (%) on the ambiguous ChangeIt video frames: multi-turn multiple choice in which the model may name a category, abstain, or pick a proactive suggestion that yields a new frame, and a trajectory counts only when it ends in the correct category; samples a first-turn guess solves for at least 25% of the evaluated models are filtered out; higher is better. Source: arxiv.org. Saturation forecast: Around 2031. 22 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | O4 Mini | 37.9 | #172 |
| 2 | Phi-4 Multimodal Instruct | 32.6 | #896 |
| 3 | InternVL3-38B | 27 | #395 |
| 4 | GPT-4.1 | 23.5 | #240 |
| 5 | GPT-5.2 | 21.7 | #105 |
| 6 | InternVL3-78B | 17.4 | #345 |
| 7 | Qwen 2.5 VL 32B Instruct | 17.4 | #443 |
| 8 | InternVL3-8B | 10.1 | #606 |
| 9 | Qwen 2.5 VL 7B Instruct | 9.8 | #643 |
| 10 | Qwen 2.5 VL 72B Instruct | 9.3 | #364 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=proactivebench-changeit · How It Works · Data refreshed daily, snapshot 2026-10-11.