EIBench (Emotion Management) - Charm: leaderboard
Metric: Anchor-based outcome score x 100 (-100 to +100): the simulated user's final emotion and relation state mapped per axis to 0 at the start anchor, +1 at the success anchor and -1 at the failure anchor (clipped), averaged over the 46 Charm scenarios (building rapport with a stranger; the model speaks first); 213 held-out test scenarios, multi-turn dialogue with an LLM user simulator, Qwen3-Max as the simulator; higher is better. Source: arxiv.org. Saturation forecast: Around 2028. 15 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Sonnet 4.6 | 43.8 |
| 2 | GPT-5.4 | 41.5 |
| 3 | Qwen 3.6 Max Preview | 39.1 |
| 4 | GLM-5.1 | 37.9 |
| 5 | Kimi K2.6 | 37.3 |
| 6 | Gemini 3 Flash | 37 |
| 7 | Qwen 3 Max | 32 |
| 8 | Seed 2.0 Pro | 31.9 |
| 9 | Gemini 3.1 Pro (Preview) | 31.3 |
| 10 | Grok 4.20 | 31 |
| 11 | DeepSeek V4 Pro | 28 |
| 12 | Grok 4.20 (Reasoning) | 28 |
| 13 | Qwen 3 32B | 23.7 |
| 14 | MiniMax-M2.5 | 21.6 |
| 15 | Qwen 3 8B | 14.6 |
Interactive version: theaggregate.ai/benchmark?slug=eibench-emotion-management-charm · How It Works · Data refreshed daily, snapshot 2026-09-29.