SEA-HELM (Thai): leaderboard
AI Singapore’s SEA-HELM restricted to Thai: the mean over its natural-language understanding, generation, reasoning, instruction-following and cultural tasks in that language. SEA-HELM exists because Southeast Asian languages are where English-first models lose the most, and the per-language boards are where that shows.
Metric: Normalized Score. Source: leaderboard.sea-lion.ai. Saturation forecast: Around January 2027. 58 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3.5 27B (Reasoning) | 73.53 | #187 (Qwen 3.5 27B) |
| 2 | Qwen 3.6 27B (Reasoning) | 73.22 | #170 (Qwen 3.6 27B) |
| 3 | Gemma 4 26B A4B (IT) (Reasoning) | 72.94 | #297 (Gemma 4 26B A4B (IT)) |
| 4 | Qwen 3.6 35B A3B (Reasoning) | 72.84 | #226 (Qwen 3.6 35B A3B) |
| 5 | Qwen 3.5 122B A10B (Reasoning) | 72.75 | #173 (Qwen 3.5 122B A10B) |
| 6 | Qwen 3.5 35B A3B (Reasoning) | 71.49 | #256 (Qwen 3.5 35B A3B) |
| 7 | Qwen 3.5 9B (Reasoning) | 68.77 | #379 (Qwen 3.5 9B) |
| 8 | Qwen 3 VL 32B Instruct | 67.93 | #283 |
| 9 | Qwen 3 32B (Thinking) | 66.95 | #443 (Qwen 3 32B) |
| 10 | GPT-OSS-120B (Reasoning) | 66.88 | #341 (GPT-OSS-120B) |
| 11 | Qwen 3.5 4B (Reasoning) | 65.58 | #475 (Qwen 3.5 4B) |
| 12 | Llama-SEA-LION-v3-70B-IT | 65.53 | #402 |
| 13 | Qwen 3 14B (Reasoning) | 65.47 | #537 (Qwen 3 14B) |
| 14 | Qwen 3 8B (Thinking) | 63.37 | #674 (Qwen 3 8B) |
| 15 | Llama 3.3 70B Instruct | 62.3 | #534 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=sea-helm-thai · How It Works · Data refreshed daily, snapshot 2026-10-20.