ProactBench: leaderboard
Most LLM benchmarks score how well a model responds to explicit requests. They leave unmeasured a different conversational ability: noticing and acting on needs the user has implied but not said. We call this conversational proactivity. ProactBench decomposes it into three phase-tied types: Emergent, inference from a single disclosed anchor; Critical, synthesis across m
Metric: Overall Pass Rate (self-reported). Source: benchmarklist.com. Saturation forecast: Around June 2027. 16 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.5 | 61.5 | #31 |
| 2 | Qwen 3.5 397B A17B | 52.5 | #146 |
| 3 | Claude Opus 4.7 | 51.9 | #49 |
| 4 | Gemini 3.1 Pro (Preview) | 50.4 | #58 |
| 5 | Qwen 3.5 9B | 48 | #379 |
| 6 | Kimi K2.6 | 47.5 | #104 |
| 7 | MiMo-V2.5-Pro | 44.9 | #158 |
| 8 | DeepSeek V4 Flash | 44.3 | #137 |
| 9 | Gemini 2.5 Pro | 42.5 | #152 |
| 10 | O4 Mini | 38.6 | #180 |
| 11 | Gemini 2.5 Flash | 35.3 | #245 |
| 12 | GPT-4o | 20.5 | #343 |
| 13 | Llama 4 Maverick | 13.1 | #464 |
| 14 | Qwen 2.5 7B Instruct | 11.5 | #862 |
| 15 | Qwen 3 1.7B | 10.3 | #1196 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=proactbench · How It Works · Data refreshed daily, snapshot 2026-10-20.