Agnes-SeaLLM-8B

Sapiens AI's only open-weight model: a compact 8B LLM fine-tuned from Qwen3-8B-Base for Southeast Asian languages (Chinese, English, Indonesian, Vietnamese, Thai), with a focus on math reasoning, translation, instruction-following, and reduced hallucination on culturally specific queries. Released under Apache 2.0 with GGUF quantizations for llama.cpp / LM Studio / Ollama.

Reported: SeaExam average 75.32, M3Exam 77.47, MMLU 74.13 — a modest, regionally-focused open release from an otherwise closed-model, consumer-API-first company.