GitHub radar
Qwen3.8-Flash-Next: Qwen's next architecture — watch, don't install
This is an early draft of the architecture that will power Qwen4. Read its card to see where Qwen is headed, but you won't be able to run this model yourself.
A preview of Qwen's new internal design that reads text, images, and video together, and beats the previous Qwen generation on coding and agent benchmarks.
When it helps
Useful if you want an early look at features that will later land in smaller Qwen models. Skip it if you're looking for a model to run on your own machine — none exists here.
Pros
- Clearly beats the previous Qwen generation on coding tasks
- Stronger on long agent tasks and phone-control benchmarks
- Context window up to 1 million tokens for huge inputs
- Previews architecture ideas that will reach smaller models later
Cons
- 180 billion parameters — needs a server GPU cluster
- No lightweight or home-runnable version exists yet
- This is an experimental preview, not a ready-to-use tool
How to set it up — step by step
- 1Open your AI agent (Claude Code, Codex) in your project folder.
- 2Send the agent the text from the block below.
- 3Read its answer on how this architecture differs from Qwen3.8-27B.
- 4Note that a compact version for your own project isn't out yet.
- 5Save the model card link so you can check back later.
Text for your agent
Copy this and send it to your agent — Claude Code, Codex, any of them:
Read the Qwen3.8-Flash-Next model card at https://huggingface.co/Qwen/Qwen3.8-Flash-Next — explain what's architecturally new compared to Qwen3.8-27B and when to expect smaller local variants based on this architecture.
Server-grade model — you can't run this at home. 180 billion parameters require a GPU cluster. Think of it as news, not a laptop tool.
Open on Hugging Face▌ More finds