Alibaba / Qwen · 2026-04-02 · major
Qwen 3.6-Plus — agentic coding flagship
Alibaba's new flagship with a 1M-token native context window, hybrid linear-attention MoE architecture, and agentic coding focus. Rivals Claude Opus 4.6 on coding benchmarks at 158 tok/s.
Alibaba's new flagship language model with a million-token context window and agentic coding that rivals Opus 4.6 at nearly double the speed.
Key specs
| Context window | 1M tokens |
|---|---|
| Output speed | ~158 tok/s |
| Pricing (bailian) | ~$0.29/M input |
What is it?
Qwen 3.6-Plus is Alibaba's next-generation flagship model, announced April 2, 2026. It features a hybrid architecture combining linear attention with sparse mixture-of-experts routing, a native 1M-token context window (roughly 2,000 pages of text or an entire large codebase), and 65,536 max output tokens. Alibaba is positioning it specifically for agentic coding — tasks where the model breaks down complex programming problems, writes and tests code, and troubleshoots iteratively.
How does it work?
The hybrid linear-attention MoE architecture delivers inference speed of approximately 158 tokens per second, roughly 1.7x faster than Claude Opus 4.6 and 2x faster than GPT-5.4. The model can natively process images and generate front-end web pages from screenshots, design drafts, and text prompts. It is available through the Alibaba Cloud Bailian platform and for free preview on OpenRouter.
Why does it matter?
The combination of a 1M context window, near-frontier coding performance, and inference speed that doubles the competition makes this interesting for teams running long-horizon coding agents where both quality and throughput matter. The free OpenRouter preview lowers the barrier to try it.
Who is it for?
Developers building agentic coding pipelines, teams evaluating alternatives to Claude and GPT for code tasks.
Try it
openrouter.ai/qwen/qwen3.6-plus-preview (free preview)