Hy-MultiTurn: leaderboard
Hy-MultiTurn (Tencent Hunyuan, 2026) tests deep multi-turn understanding in Chinese on 209 long dialogues of 12 to 76 turns across six dimensions. The score is the importance-weighted percent of checklist items the final response satisfies, averaged over three rolls and three LLM judges (a deterministic checker for precise execution); higher is better.
Metric: Importance-weighted score (%): share of the importance-weighted checklist items a response satisfies, 209 long Chinese multi-turn dialogues (12 to 76 turns), three rolls per task averaged, three LLM judges (GPT-5.2, Qwen 3.6, DeepSeek-V4 Pro) averaged, a deterministic checker for precise execution; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 22 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.5 (High) | 77.2 | #49 (GPT-5.5) |
| 2 | Grok 4.5 (High) | 75.8 | #82 (Grok 4.5) |
| 3 | Kimi K3 (Max) | 74.2 | #54 (Kimi K3) |
| 4 | GPT-5.4 (High) | 73.3 | #123 (GPT-5.4) |
| 5 | Gemini 3.1 Pro (Preview) (High) | 71.7 | #94 (Gemini 3.1 Pro (Preview)) |
| 6 | Claude Opus 4.6 (High) | 71.6 | #102 (Claude Opus 4.6) |
| 7 | Claude Opus 4.7 (High) | 70.8 | #79 (Claude Opus 4.7) |
| 8 | Hy3 (High) | 67.2 | #195 (Hy3) |
| 9 | Kimi K2.6 | 67 | #178 |
| 10 | Doubao-Seed-2.1-Pro (High) | 62.9 | |
| 11 | Kimi K2.5 (Thinking) | 60.1 | #256 (Kimi K2.5) |
| 12 | Qwen 3.6 Plus (Thinking) | 58.7 | #222 (Qwen 3.6 Plus) |
| 13 | Kimi K2.6 (Non-reasoning) | 56.4 | #178 (Kimi K2.6) |
| 14 | Hy3-preview (High) | 55.7 | #316 (Hy3-preview) |
| 15 | Kimi K2.5 (Non-reasoning) | 53 | #256 (Kimi K2.5) |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=hy-multiturn · How It Works · Data refreshed daily, snapshot 2026-11-13.