SEA-LION
Singapore's national LLM family built for Southeast Asian languages and cultures, from Burmese and Khmer to Tamil and Vietnamese.
About
SEA-LION (Southeast Asian Languages in One Network) is AI Singapore's flagship open model family, funded under Singapore's national AI program. It targets the region's languages that global models handle poorly: Indonesian, Malay, Thai, Vietnamese, Filipino, Burmese, Khmer, Lao, Tamil, and Mandarin, among others. A proprietary SEABPETokenizer with a 256K vocabulary keeps tokenization efficient for these scripts, and training draws on over a trillion tokens of regional text.
The family has evolved rapidly, from the original Llama-based v1 through v3 to v4 variants built on Gemma, Qwen, and NVIDIA Nemotron bases, plus vision-language and safety-focused SEA-Guard siblings. AI Singapore reports top open-model rankings on the SEA-HELM regional benchmark. Licensing varies by variant: early releases are Apache-2.0, while newer ones inherit their base models' terms. A public playground and API key manager lower the barrier for regional developers.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Author-reported (SEA-HELM leaderboard): top open-model rankings for Southeast Asian tasks, competitive with larger proprietary models on regional sentiment and translation tasks
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 7/10
Scored 7/10 (verified tier, metrics-v1). Evidence: AI Singapore national-program release with SEA-HELM benchmark leadership confirmed via the project GitHub and regional press; actively maintained through v4.8 (2025) across Gemma, Qwen, and Nemotron bases. Strengths: unique regional coverage with a purpose-built tokenizer, strong institutional backing, playground plus API access. Docked for: (1) licensing is a patchwork across variants, requiring per-model checks; (2) smaller global community than the major open families. License: varies by variant (Apache-2.0 for early releases; base-model terms for v4+).
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 7.0/10
- Maintenance · recency, releases, issue load
- 8.0/10
- Openness · license + docs
- 7.0/10
- Manual review · sanity read + red-flag check
- 7.0/10
- · AI Singapore national-program release (Dec 2023), actively maintained through v4.8 (2025)
- · SEA-HELM regional benchmark leadership per author reports
- · Proprietary SEABPETokenizer (256K vocab); playground + API key manager at sea-lion.ai