SEA LLM Leaderboard - SeaBench - Thai (Private): leaderboard
SeaBench private set (SeaLLMs' in-house held-out questions) score in Thai: open-ended instruction-following and multi-turn conversation questions, each response rated 1-10 by GPT-4o-2024-08-06. Higher is better; from DAMO-NLP-SG's LLM Leaderboard for SEA.
Metric: Private Judge Score (1-10). Source: huggingface.co. Saturation forecast: Estimated already saturated. 41 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4.1 | 8.89 | #258 |
| 2 | DeepSeek V3 (0324) | 8.69 | #349 |
| 3 | Claude 3.7 Sonnet (20250219) | 8.67 | #209 |
| 4 | Gemini 2.0 Flash | 8.55 | #351 |
| 5 | GPT-4o (2024-08-06) | 8.53 | #343 |
| 6 | Claude 3.5 Sonnet (20241022) | 8.52 | #302 |
| 7 | Gemini 1.5 Pro | 8.52 | #454 |
| 8 | GPT-4.1 Mini | 8.48 | #363 |
| 9 | Gemma 3 27B (IT) | 8.36 | #528 |
| 10 | Gemini 2.0 Flash Lite | 8.15 | #455 |
| 11 | GPT-4o Mini (2024-07-18) | 8.02 | #568 |
| 12 | Gemma 3 12B (IT) | 7.96 | #672 |
| 13 | Gemini 1.5 Flash | 7.87 | #620 |
| 14 | Sailor2-20B-Chat | 7.72 | #589 |
| 15 | Qwen 2.5 72B Instruct | 7.65 | #469 |
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-seabench-thai-private · How It Works · Data refreshed daily, snapshot 2026-11-05.