LiveReleased Mar 2025
ChindaLLM
Chinda LLM 4B is a Thai-first language model with step-by-step thinking-mode reasoning, adapted and tuned from Qwen3-4B.
4B parameters40k token contextApache-2.0 open weights
By the numbers
Benchmark scores vs Typhoon 2.1 Gemma3 4B
ChindaLLM Typhoon 2.1 4B
MATH500 (EN)
LiveCodeBench (EN)
OpenThaiEval (TH)
Average
Scores rescaled to 100. Source: OpenThaiGPT benchmark suite. Higher is better.
How it works
Thai-first
Trained and tuned for Thai understanding and generation, with strong English too.
Thinking mode
Explicit step-by-step reasoning for math, code and logic problems.
Runs anywhere
A 4B model fits on one ~8 GB GPU, or runs locally via Ollama, LM Studio or OpenWebUI.
Open and compatible
Open weights under Apache-2.0, served through an OpenAI-compatible API.
Where teams use it
- Thai-language assistants, chat and document question-answering.
- Reasoning over Thai math, logic and code tasks.
- On-prem or private deployments where the weights and data stay in your control.
In plain words
Sizes4 billion parameters
LanguagesThai and English
ContextUp to 40,960 tokens, enough for long Thai documents
ReasoningLeads a strong 4B Thai model on average0.569 vs 0.414 average, OpenThaiGPT suite
Built onQwen3-4B, with iApp Thai tuning
LicenseApache-2.0, open weights
To runOne ~8 GB GPU, on-prem or cloud