Top AI Product

A 480-point HN thread called Qwen 3.6 27B “the sweet spot for local development.” That’s not hype — it’s the first time this year an open-weight local model felt genuinely good enough.

What it actually is

Qwen 3.6 27B is a dense, open-weight coding model from Alibaba’s Qwen team, Apache 2.0, dropped April 22. Dense matters here: all 27B parameters fire on every pass, no MoE routing. It scores 77.2 on SWE-bench Verified — within reach of Claude 4.5 Opus’s 80.9, and it edges past Alibaba’s own 397B MoE on agentic coding. 256K context, 201 languages.

The real trick is the size. 27B sits exactly where capability meets a single high-end card — people report ~50 tok/s on an RTX 5090, ~30 on an M5 Mac. No cloud, no token bill, your code never leaves the machine.

Running it

Grab the weights from Hugging Face or ModelScope and serve locally via llama.cpp or LM Studio, or hit the Qwen API if you’d rather not host. Typical use: repo-level agentic coding, autocomplete, frontend work — the stuff you’d normally pay Claude or GPT to do, now offline and free.

You Might Also Like

• Alibaba Qwen 3 5 Just Dropped and it Brought 10 Million Milk Teas With it
• Cursor Composer 2 Takes on Anthropic and Openai With a 0 50 m Token Coding Model and the Benchmarks Back it up
• Alibaba Qwen Smart Glasses g1 s1 275 ai Glasses With Swappable Batteries and a Qwen api Backend
• Qwen Image 2 0 Just Dropped and i Honestly Wasnt Expecting This
• Runway gen 4 5 Just Took the top Spot in ai Video and its not Even Close