Hy-MultiTurn - Object Localization: leaderboard

Metric: Importance-weighted score (%) on the 35 object localization dialogues, 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: Estimated already saturated. 22 models tracked.

Top models

#ModelScore
1Grok 4.5 (High)95.2
2Kimi K3 (Max)92.1
3Claude Opus 4.6 (High)92
4GPT-5.5 (High)90.7
5Claude Opus 4.7 (High)90.7
6GPT-5.4 (High)87.9
7Gemini 3.1 Pro (Preview) (High)87.7
8Hy3 (High)86.1
9Doubao-Seed-2.1-Pro (High)85.5
10Kimi K2.682.3
11Seed 2.0 Pro (Thinking)79.9
12Kimi K2.5 (Thinking)78.2
13Hy3-preview (High)74.5
14Qwen 3.6 Plus (Thinking)72.3
15Kimi K2.5 (Non-reasoning)60.6

Interactive version: theaggregate.ai/benchmark?slug=hy-multiturn-object-localization · How It Works · Data refreshed daily, snapshot 2026-09-29.