CarryOnBench - Utility (Benign Clarifications): leaderboard

CarryOnBench (2026) starts from seemingly harmful requests that have a benign purpose and checks whether a model gives the benign information a user needs, and how harmful its answers get, as simulated users clarify or push back. Share of the benign information checklist fulfilled in multi-turn conversations where the user clarifies the benign intent, counted over the whole conversation, in percent; higher is better.

Metric: Ben-Util (%): share of the benign information-need checklist (about 18 atomic yes/no items per query, built from five models' compliant answers) that the responses fulfil, each item judged separately by Gemini-2.5-Flash, over multi-turn conversations of 4 to 12 turns in which at least one user turn clarifies the benign intent, conversation-level, on CarryOnBench (398 seemingly harmful SORRY-Bench queries with a benign underlying intent; user turns simulated by Gemini-3-Flash); higher is better. Source: arxiv.org. Saturation forecast: Not forecast. 14 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Qwen 3 Next 80B A3B Instruct75.2#778
2GPT-5 Mini74.8#364
3Qwen 3 32B73.8#972
4Qwen 3 235B A22B71.2#628
5DeepSeek V3.166.3#511
6GPT-5.465.4#135
7OLMo 3.1 32B Instruct64.4#2445
8OLMo 3 7B Instruct61.3#6953
9Gemini 3.1 Pro (Preview)57.4#99
10GPT-OSS-120B52.5#704
11Claude Sonnet 451.3#410
12Claude Haiku 4.549.9#583
13Mixtral 8x7B Instruct45.1#4696
14Llama 4 Maverick37#1110

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

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