Qwen 3.6 Is Live — Three New Models, Same Prices

April 30, 2026

Alibaba shipped the Qwen 3.6 family and three of them are in your model picker today. They're better at reasoning, they all handle images and video, and the new Flash tier reads up to a million tokens in one pass. Best part: nothing got more expensive.

The three new options

• Qwen 3.6 Flash — fast, multimodal, and a 1M-token context window (roughly 750,000 words in one shot). Great when you've got a huge document and don't want to chop it up. Budget pricing.
• Qwen 3.6 27B — a strong everyday reasoner with built-in thinking mode and a 256K context window. Same price as the 3.5 version it replaces.
• Qwen 3.6 Max Preview — Alibaba's new flagship, with a 256K context window. Built for long, multi-step coding and tool-use jobs where the model needs to keep its train of thought across many turns. Same price as the 3.5 flagship it replaces.

If your sessions were pinned to an old Qwen 3.5 model, they roll forward to the right 3.6 model on your next message. You don't need to do anything.

When you'd actually reach for one

• You've got a 600K-token research dump and don't want to summarize the back half away. Send it to Flash in one shot.
• You're running a long agentic coding loop with lots of tool calls. Max Preview is the one that holds its reasoning across the whole loop instead of starting over each turn.
• For everything in between, 27B is the workhorse — auto-routing will land you there for most prompts that used to go to Qwen 3.5.

Try it

• "Refactor this module end-to-end and walk me through each step."
• "Pull every market-sizing claim out of my full research corpus."
• "Run this engineering brief through swarm and synthesize the strongest take."

Heads up

• The matching Qwen 3.5 versions are retired from the picker. Auto-routing won't pick them either.
• Local Qwen 3.5 (for private deployments and Apple Silicon) keeps working until the 3.6 local builds are validated.
• The open-weight 35B variant isn't in the cloud picker yet — its tool-calling isn't reliable enough. The Apple Silicon build still works.