Qwen3-Coder is a new 480B MoE open model (35B active) by the Qwen team, built for agentic coding. It achieves SOTA results on benchmarks like SWE-bench, supports up to 1M context, and comes with an open-source CLI tool, Qwen Code.
The Qwen team has released Qwen3-Coder, a new open-source model aimed squarely at agentic coding.
It's a massive 480B Mixture-of-Experts model (35B active) achieving top results for open models on tough agentic benchmarks like SWE-bench. It also has a huge context window for working with entire codebases.
To help developers get started, they've also open-sourced a CLI tool called Qwen Code (forked from Gemini CLI), and you can use the model with other tools like Cline. The model is available now via the official Alibaba Cloud API, and I hear OpenRouter will have an API soon, with Unsloth also working on a version for local runs.
Qwen3-Coder looks like a serious leap forward for agentic coding — massive context, strong benchmarks, and open tooling make it a standout for devs and researchers alike. Congrats on the launch 🚀
Massive leap forward! A 480B MoE model that's open, tuned for agentic coding, and supports 1M context? Qwen3-Coder is setting a new bar. Can’t wait to try it out in a real coding workflow.
Replies
Hi everyone!
The Qwen team has released Qwen3-Coder, a new open-source model aimed squarely at agentic coding.
It's a massive 480B Mixture-of-Experts model (35B active) achieving top results for open models on tough agentic benchmarks like SWE-bench. It also has a huge context window for working with entire codebases.
To help developers get started, they've also open-sourced a CLI tool called Qwen Code (forked from Gemini CLI), and you can use the model with other tools like Cline. The model is available now via the official Alibaba Cloud API, and I hear OpenRouter will have an API soon, with Unsloth also working on a version for local runs.
BestPage.ai
Whoa, Qwen3 sounds wild—love how it lets you actually choose between deep thinking and speed. That’s such a gamechanger for real work, tbh!
Linkinize
Qwen3-Coder looks like a serious leap forward for agentic coding — massive context, strong benchmarks, and open tooling make it a standout for devs and researchers alike. Congrats on the launch 🚀
Tough Tongue AI
The CLI tool and the visual presentation of this launch give it a polished, developer-friendly vibe. Really makes me want to try it.
Seveum
very cool model, maybe we could use to help people develop android / iOS apps on our platform too :)
Massive leap forward! A 480B MoE model that's open, tuned for agentic coding, and supports 1M context? Qwen3-Coder is setting a new bar. Can’t wait to try it out in a real coding workflow.
I'd love to have a try
This is wild! Competing directly with Claude is a bold move, and I'd follow Qwen3 closely to see how it evolves!
The Qwen team always delivers. What a great model! Super grateful and can't wait for the new thinking model of the 235B-A22B. 🙏