CarryOnBench - Utility (Turn 1): leaderboard
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 the first response to the bare query, 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.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-5 Mini | 37.6 |
| 2 | Qwen 3 32B | 32 |
| 3 | GPT-5.4 | 31.4 |
| 4 | Qwen 3 Next 80B A3B Instruct | 30.6 |
| 5 | DeepSeek V3.1 | 29.6 |
| 6 | Olmo 3.1 32B Instruct | 23.5 |
| 7 | OLMo 3 7B Instruct | 23.3 |
| 8 | Gemini 3.1 Pro (Preview) | 23.2 |
| 9 | Claude Sonnet 4 | 22.7 |
| 10 | Qwen 3 235B A22B | 20.3 |
| 11 | Claude Haiku 4.5 | 20.2 |
| 12 | Mixtral 8x7B Instruct (v0.1) | 20 |
| 13 | Llama 4 Maverick | 13.2 |
| 14 | GPT-OSS-120B | 10.5 |
Interactive version: theaggregate.ai/benchmark?slug=carryonbench-utility-turn-1 · How It Works · Data refreshed daily, snapshot 2026-10-07.