Sailor2-20B-Chat: benchmark results
Sea AI Lab's Sailor2-20B-Chat, a Southeast-Asia-focused multilingual chat model continually pretrained from Qwen 2.5 14B to 20B on 500B tokens covering 15 SEA languages. Provider: Sea AI Lab. Released 2025-02-18. Access: Open.
Aggregate Rating 1558 ± 20, rank #593 of 1661 rated models, from 49 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Thai LLM - Coding | 7.82 | Rating (0-10) | 93.7 |
| Thai LLM NLU - xcopa_tha_seacrowd_qa | 92.8 | Accuracy (%) | 92.8 |
| FilBench - Classical NLP | 89.03 | Category Score (%) | 87.5 |
| SEA LLM Leaderboard - SeaExam - Indonesian (Private) | 77.2 | Private Accuracy (%) | 85.6 |
| SEA LLM Leaderboard - SeaExam - Vietnamese (Public) | 74.8 | Public Accuracy (%) | 85.6 |
| Thai LLM - Social Science | 8.4 | Rating (0-10) | 85.2 |
| SEA LLM Leaderboard - SeaExam (Public) | 71.6 | Public Average Score (%) | 84.4 |
| Thai LLM - Roleplay | 7.6 | Rating (0-10) | 83.8 |
| Thai LLM - Math | 6.7 | Rating (0-10) | 83.1 |
| Thai LLM - STEM | 7 | Rating (0-10) | 83.1 |
| Thai LLM MC - m3exam_tha_seacrowd_qa | 60.89 | Accuracy (%) | 82.6 |
| Thai LLM NLU - belebele_tha_thai_seacrowd_qa | 85.78 | Accuracy (%) | 82.6 |
Interactive version: theaggregate.ai/model?slug=sailor2-20b-chat · How It Works · Data refreshed daily, snapshot 2026-10-23.