Qwen3
Think Deeper or Act Faster
5.0•15 reviews•2.2K followers
Think Deeper or Act Faster
5.0•15 reviews•2.2K followers
Qwen3 is the large language model series developed by Qwen team, Alibaba Cloud. - QwenLM/Qwen3
This is the 19th launch from Qwen3. View more
Qwen3.6-27B
Launching today
Qwen3.6-27B is a fully open-source dense model that punches way above its weight. Surpassing the previous 397B MoE flagship in agentic coding, it supports multimodal reasoning and thinking modes while remaining perfectly sized for local self-hosting.
Free
Launch Team
DeepSeek
Flowtica Scribe
Hi everyone!
This is a dense model, not an MoE, and that matters.
Dense models often show unusually strong intelligence density for their size, and Qwen3.6-27B is a very good example of that. At 27B, it already pushes past Qwen3.6-35B-A3B on a number of key coding and reasoning tasks, and more importantly, it beats the previous open-source flagship Qwen3.5-397B-A17B across all major coding benchmarks. That is a pretty serious result for a dense checkpoint at this scale.
And 27B is also just a very sweet open-source size. It is not so large that normal users or small teams are locked out of deployment, but it is not small either — it still leaves a lot of headroom for real capability.
In the Qwen3.6 era, this has a very real chance of becoming their most popular open dense model.
we're almost there. if that long term coding gets bumped up and reaches opus 4.5 levels we're looking at something serious. i reckon in about 6-9 months they are there or beyond that and approaching opus 4.6 levels. at that rate, for coding, running that locally on your hardware, it's hard to justify picking a frontier cloud model.