CUE-Bench - Pragmatic Intent: leaderboard
CUE-Bench (Huazhong University of Science and Technology, 2026) evaluates affective understanding of Chinese discourse beyond surface sentiment, using contextual instances across three connected tasks. Accuracy on pragmatic intent classification under zero-shot direct prompting, in percent, higher is better.
Metric: Accuracy (%; pragmatic intent classification; over CUE-Bench's Chinese-discourse instances with context; zero-shot direct prompting). Source: arxiv.org. Saturation forecast: Around 2030. 6 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | DeepSeek V4 Flash | 32.5 | #254 |
| 2 | GLM-5.1 | 31.2 | #175 |
| 3 | GPT-4o Mini | 28.1 | #1627 |
| 4 | Llama 4 Maverick | 27.3 | #1105 |
| 5 | Llama 3.1 8B | 23.2 | #4900 |
| 6 | Qwen 3 8B | 19.6 | #1838 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=cue-bench-pragmatic-intent · How It Works · Data refreshed daily, snapshot 2026-11-12.