SEA-HELM (Malay) - Toxicity Detection: leaderboard
Toxicity detection: classifying whether Malay social-media text is toxic or abusive. Normalized 0-100 score from AI Singapore's SEA-HELM leaderboard: (score - random baseline) / (maximum - baseline), floored at 0 and averaged over eight runs; higher is better.
Metric: Normalized Score. Source: leaderboard.sea-lion.ai. 58 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Llama 3.3 70B Instruct | 17.51 | #541 |
| 2 | Qwen 3 VL 8B Instruct | 16.84 | #419 |
| 3 | Apertus-70B-Instruct-2509 | 16.57 | #819 |
| 4 | Llama 3.1 8B Instruct | 14.82 | #1017 |
| 5 | MERaLiON-2-10B | 14.61 | #600 |
| 6 | Gemma 3 27B (IT) | 14.41 | #528 |
| 7 | Llama-SEA-LION-v3-8B-IT | 14.39 | #800 |
| 8 | Qwen 3 VL 32B Instruct | 14.1 | #290 |
| 9 | Gemma 3 12B (IT) | 13.81 | #672 |
| 10 | Llama-SEA-LION-v3-70B-IT | 13.68 | #407 |
| 11 | Llama 4 Scout Instruct | 13.64 | #547 |
| 12 | Qwen 3 VL 4B Instruct | 12.9 | #510 |
| 13 | Gemma 4 26B A4B (IT) (Reasoning) | 12.08 | #293 (Gemma 4 26B A4B (IT)) |
| 14 | Mistral Small 4 | 12.03 | #522 |
| 15 | Qwen 3.5 4B (Reasoning) | 11.82 | #514 (Qwen 3.5 4B) |
Interactive version: theaggregate.ai/benchmark?slug=sea-helm-malay-toxicity-detection · How It Works · Data refreshed daily, snapshot 2026-11-05.