EIBench (Emotion Management) - Charm: leaderboard
EIBench (2026) is a simulator-based benchmark for interactive emotion management: an LLM plays the user in a multi-turn scenario and updates an emotion and relation state after every reply, and the final state is scored from -100 to +100 relative to scenario-specific start, success and failure anchors. This board averages the 46 Charm scenarios, where the model builds rapport with a stranger, with Qwen3-Max as the simulator; higher is better.
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 October 2027. 15 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Sonnet 4.6 | 43.8 | #161 |
| 2 | GPT-5.4 | 41.5 | #123 |
| 3 | Qwen 3.6 Max Preview | 39.1 | #159 |
| 4 | GLM-5.1 | 37.9 | #179 |
| 5 | Kimi K2.6 | 37.3 | #178 |
| 6 | Gemini 3 Flash | 37 | #164 |
| 7 | Qwen 3 Max | 32 | #403 |
| 8 | Seed 2.0 Pro | 31.9 | #211 |
| 9 | Gemini 3.1 Pro (Preview) | 31.3 | #94 |
| 10 | Grok 4.20 | 31 | #297 |
| 11 | DeepSeek V4 Pro | 28 | #188 |
| 12 | Grok 4.20 (Reasoning) | 28 | #297 (Grok 4.20) |
| 13 | Qwen 3 32B | 23.7 | #960 |
| 14 | MiniMax-M2.5 | 21.6 | #582 |
| 15 | Qwen 3 8B | 14.6 | #1830 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=eibench-emotion-management-charm · How It Works · Data refreshed daily, snapshot 2026-11-13.