Thai LLM NLU - belebele_tha_thai_seacrowd_qa: leaderboard
The Thai LLM Leaderboard, run as an open results dataset by the Thai LLM community. Natural-language understanding accuracy on Belebele’s Thai reading-comprehension split (SEACrowd harness).
Metric: Accuracy (%). Source: huggingface.co. Saturation forecast: Estimated already saturated. 70 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude 3.5 Sonnet (20240620) | 90.22 | #299 |
| 2 | GPT-4o (2024-05-13) | 89.78 | #325 |
| 3 | Llama 3.1 Nemotron 70B Instruct | 88.33 | #554 |
| 4 | Qwen 2.5 32B Instruct | 88.22 | #498 |
| 5 | Qwen 2.5 72B Instruct | 87.67 | #454 |
| 6 | Llama 3.1 70B Instruct | 87.56 | #561 |
| 7 | Llama 3.1 70B Cpt Sea Lionv3 Instruct | 87.22 | #548 |
| 8 | Llama 3.3 70B Instruct | 86.67 | #534 |
| 9 | Qwen 2 72B Instruct | 86.11 | #576 |
| 10 | Sailor2-20B-Chat | 85.78 | #590 |
| 11 | Qwen 2.5 14B Instruct | 85.44 | #647 |
| 12 | gemma2-9B-cpt-sea-lionv3-instruct | 85 | #705 |
| 13 | GPT-4o Mini (2024-07-18) | 84.78 | #558 |
| 14 | QwQ 32B-Preview | 84 | #651 |
| 15 | Gemma 2 9B (IT) | 83.78 | #788 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=thai-llm-nlu-belebele-tha-thai-seacrowd-qa · How It Works · Data refreshed daily, snapshot 2026-10-20.