EYT-Bench (Nemotron-USA) - Latent Intent: leaderboard
Metric: Per-turn latent-intent accuracy (%; the target model first predicts the user latent-intent label from 8 classes, scored against the simulator gold label, before generating its reply; 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); Nemotron-Personas-USA pool (100 dialogues; structured demographic personas)). Source: arxiv.org. Saturation forecast: Around July 2027. 17 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | DeepSeek V4 Pro | 57.6 |
| 2 | GPT-5.5 | 49.2 |
| 3 | Gemini 3.1 Pro (Preview) (Thinking) | 46.1 |
| 4 | DeepSeek V4 Flash | 39.8 |
| 5 | Qwen 3.5 397B A17B | 36.3 |
| 6 | Claude Opus 4.7 | 34.2 |
| 7 | Qwen 3.5 27B | 28.5 |
| 8 | Claude Sonnet 4.6 | 27 |
| 9 | Qwen 3.5 35B A3B | 25.7 |
| 10 | Seed 2.0 Pro | 22.5 |
| 11 | Seed 2.0 Lite | 20.9 |
| 12 | Seed 2.0 Mini | 16 |
Interactive version: theaggregate.ai/benchmark?slug=eyt-bench-nemotron-usa-latent-intent · How It Works · Data refreshed daily, snapshot 2026-09-29.