HELM SeaHELM - ViHSD: leaderboard
HELM SeaHELM Southeast-Asian-languages evaluation: Vietnamese hate-speech detection.
Metric: Macro F1 score. Source: crfm.stanford.edu. Saturation forecast: Estimated already saturated. 21 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4o (2024-11-20) | 61.63 | #381 |
| 2 | Claude 3.7 Sonnet (20250219) | 59.89 | #201 |
| 3 | Qwen 2 72B Instruct | 54.54 | #576 |
| 4 | Qwen 2.5 72B Instruct | 53.98 | #454 |
| 5 | Gemini 2.0 Flash | 51.19 | #342 |
| 6 | Llama 3.3 70B Instruct | 50.54 | #534 |
| 7 | Llama 3.1 70B Instruct | 48.8 | #561 |
| 8 | Llama 3.1 8B CPT SEA LIONv3 Instruct | 47.91 | #810 |
| 9 | Gemini 2.0 Flash Lite | 46.06 | #450 |
| 10 | Llama 4 Scout Instruct | 45.15 | #539 |
| 11 | Llama 3.1 405B Instruct | 42.01 | #457 |
| 12 | GPT-4o Mini (2024-07-18) | 41.19 | #558 |
| 13 | Llama 4 Maverick Instruct FP8 | 39.41 | #414 |
| 14 | Llama 3.2 90B Vision Instruct | 38.29 | #622 |
| 15 | Gemma 2 27B (IT) | 25.62 | #645 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=helm-seahelm-vihsd · How It Works · Data refreshed daily, snapshot 2026-10-20.