SuiChat-CN - Few-Shot: leaderboard
Metric: Macro F1 (%) of suicide-risk level classification of Chinese group-chat topics into six levels (L0 to L5) on the SuiChat-CN test set, macro F1 over the levels, few-shot prompting with labeled examples; higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 44 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Seed-OSS-36B-Instruct | 77.3 |
| 2 | Gemini 2.5 Pro | 76.4 |
| 3 | Ring-flash-2.0 | 76 |
| 4 | DeepSeek V3 | 75.1 |
| 5 | DeepSeek R1 | 74.1 |
| 6 | Qwen Max | 73.6 |
| 7 | Kimi K2.5 | 73.4 |
| 8 | Qwen 3.5 Plus | 73.3 |
| 9 | Kimi K2 | 72.8 |
| 10 | Qwen 3.5 397B A17B | 72.4 |
| 11 | GPT-5 | 72 |
| 12 | Kimi K2.6 | 71.9 |
| 13 | Llama 3.3 70B Instruct | 71.2 |
| 14 | GLM-5.1 | 71.1 |
| 15 | Qwen 3 235B A22B | 70.8 |
Interactive version: theaggregate.ai/benchmark?slug=suichat-cn-few-shot · How It Works · Data refreshed daily, snapshot 2026-10-07.