SEA LLM Leaderboard - SeaExam - Thai (Private): leaderboard
SeaExam private set (SeaLLMs' in-house held-out questions) accuracy in Thai: multiple-choice questions from real human exams testing world knowledge and reasoning, 3-shot for open models and zero-shot for API models. 0-100, higher is better; from DAMO-NLP-SG's LLM Leaderboard for SEA.
Metric: Private Accuracy (%). Source: huggingface.co. Saturation forecast: Estimated already saturated. 91 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.5 Pro (Preview 03-25) | 94.2 | #126 |
| 2 | Claude 3.5 Sonnet (20241022) | 76 | #302 |
| 3 | Claude 3.7 Sonnet (20250219) | 74.1 | #209 |
| 4 | Gemini 1.5 Pro | 73.9 | #454 |
| 5 | DeepSeek V3 (0324) | 73.7 | #349 |
| 6 | Gemini 2.0 Flash | 73.2 | #351 |
| 7 | Llama 4 Maverick Instruct | 71.8 | #467 |
| 8 | Qwen 2.5 72B | 71.3 | #512 |
| 9 | Qwen 2 72B | 69.7 | #515 |
| 10 | GPT-4o (2024-08-06) | 69.5 | #343 |
| 11 | Qwen 2.5 72B Instruct | 68.8 | #469 |
| 12 | GPT-4.1 | 68.5 | #258 |
| 13 | Gemini 2.0 Flash Lite | 67.4 | #455 |
| 14 | Gemini 1.5 Flash | 65.7 | #620 |
| 15 | Qwen 2.5 14B | 65 | #716 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=sea-llm-leaderboard-seaexam-thai-private · How It Works · Data refreshed daily, snapshot 2026-11-05.