Qwen2.5-3B-Instruct
Qwen/Qwen2.5-3B-Instruct · Apache-2.0
A 3B all-rounder with unexpectedly good coding and 29+ languages.
Qwen2.5-3B-Instruct balances reasoning, coding and structured output, and is a popular base for fine-tunes and quantised edge deployments.
Score
Setup at a glance
pip install transformers Deep dive
Qwen2.5-3B-Instruct is the small end of the Qwen2.5 instruct family, and it behaves like a bigger model: significantly more knowledge than its predecessor, much better instruction following, and long-form generation that holds together.
It supports 128K-token context with up to 8K tokens of generation, handles 29+ languages, and is notably strong at coding for its size — the combination that makes it a popular fine-tuning base.
The quickstart is the standard transformers path, and the Qwen docs add benchmark and speed-test tooling for deployment planning.