EYT-Bench (PersonaMem-v2) - Final Intent Completion: leaderboard
EYT-Bench (Enjoy Your Talk, 2026) evaluates LLMs as multi-turn conversational partners with a persona-grounded user simulator and an independent LLM judge; each turn the target model predicts the user state before replying. Share of dialogues in which the judge finds the user final intent reached, in percent on the PersonaMem-v2 pool of free-text personas distilled from real conversations; higher is better.
Metric: Final-intent completion rate (%; share of dialogues in which the judge, reading the full transcript, decides the assistant helped the user reach the stated final intent; multi-turn dialogues of up to 10 turns with a persona-grounded Gemini 3.1 Pro user simulator and a Gemini 3.1 Pro judge (reasoning effort high, temperature 0); PersonaMem-v2 pool (100 dialogues; personas are free-text paragraphs distilled from long real user-assistant interactions)). Source: arxiv.org. 17 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Seed 2.0 Lite | 88.6 | #289 |
| 2 | Seed 2.0 Mini | 87.9 | #508 |
| 3 | DeepSeek V4 Flash | 87.5 | #258 |
| 4 | Seed 2.0 Pro | 86.2 | #211 |
| 5 | Qwen 3.5 35B A3B | 83.3 | #482 |
| 6 | Qwen 3.5 397B A17B | 81.2 | #246 |
| 7 | Qwen 3.5 27B | 80.9 | #359 |
| 8 | Gemini 3.1 Pro (Preview) (Thinking) | 80 | #94 (Gemini 3.1 Pro (Preview)) |
| 9 | DeepSeek V4 Pro | 78.1 | #188 |
| 10 | Claude Opus 4.7 | 69 | #79 |
| 11 | Claude Sonnet 4.6 | 66 | #161 |
| 12 | GPT-5.5 | 52.5 | #49 |
Interactive version: theaggregate.ai/benchmark?slug=eyt-bench-personamem-v2-final-intent-completion · How It Works · Data refreshed daily, snapshot 2026-11-13.