Introducing Jan-Code-4B 💻 A compact coding model tuned for practical day-to-day tasks. Generation, refactors, debugging, tests — all runnable locally in Jan. Download Jan: Model:

As always, thanks to the @Alibaba_Qwen team for the base model behind Jan-v3-4B and to @ggerganov for llama.cpp 💛

Local first coding models are getting genuinely practical. Great to see smaller checkpoints focused on real workflows instead of benchmark theater.

You keep surprising me with new amazing features and models!

4 😎

Generation, refactors, debugging, tests — all runnable locally in Jan. Download Jan: … q1 2026 @alexdolbun, in love with small quantized RLed models for software pipelines!! GJ!

What is problem?

You can also run Jan-Code-4B on Android using the Apex Creator Local AI app: Great to see more ways to run local models on mobile devices 👏

cannot wait to see Jan V4 based on the new 3.5!

Local inference is the final frontier for developer privacy and reduced latency. Tuning compact 4B models for specific logic tasks is far more efficient than over-relying on monolithic cloud APIs. Lean engineering wins. 💻🔒

Seems like a good model, can't wait to test.

Cool bro, I need it

@victormustar Dm?

MLX available at jedisct1/Jan-code-4b-mlx . Works pretty well with !

Thanks for your work!

@grok sitesinden mi bilgisayarıma indirip kurcam?

Nice! Thank You.. And I hate to ask, but when Qwen3.5-9B? 😬

Y'all planning on dropping any new models soon?

👀

banger good job

4B is a sweet spot for running on laptops without fans going full jet engine. Curious how it handles multi-file refactors where you need context across imports and dependencies, that's where small models usually start to fall apart.

Feeling ancient with those manual codes? Try our new kid on the block – Jan-Code-4B!

Please introduce the ability to manually change the embedding model to a different one! ...e.g., to bge-m3. This is important for other languages ​​(e.g., Polish)

Perfect, thanks for sharing, can we use opencode with Jan?