ProactBench - Emergent: leaderboard

Pass rate on the Emergent triggers of ProactBench (2026), turns 1 to 3 of curated multi-turn dialogues where the assistant must infer an unstated need from a single disclosed detail; a GPT-5.4 judge grades each response Pass, Partial or Fail against a rubric written before the model answers. Percent; higher is better.

Metric: Pass rate (%) on the Emergent triggers (turns 1-3: inferring an unstated need from a single disclosed anchor) over the 198 curated dialogues of ProactBench (about 624 rubric-scored trigger points: 201 Emergent, 232 Critical, 191 Recovery), each trigger judged Pass, Partial or Fail by a GPT-5.4 judge against a rubric written before the model answers; dialogue histories are fixed from a curation run and the model regenerates only at the triggers, sampled once at temperature 0.7; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 16 models tracked. Updated 7 Oct 2026.

Top models

#ModelSettingScoreOverall rank
1Qwen 3.5 397B A17B Instruct72.9#285
2Qwen 3.5 9B Instruct72.8#678
3Claude Opus 4.772.1#85
4Gemini 3.1 Pro (Preview)70.2#99
5GPT-5.568.7#54
6Kimi K2.667.4#192
7Gemini 2.5 ProMedium66.2#276
8MiMo-V2.5-Pro65.7#284
9DeepSeek V4 Flash61.7#267
10Gemini 2.5 Flash60.7#463
11O4 Mini (2025-04-16)52.7#370
12GPT-4o (2024-11-20)33.8#859
13ProactBench Llama-3.2-8B (checkpoint unspecified)25.9
14Llama 4 Maverick22.4#1110
15Qwen 2.5 7B Instruct21.9#2723

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=proactbench-emergent · How It Works · Data refreshed daily, snapshot 2026-11-21.