Alibaba teases Qwen4 and opens a Qwen4-Max preview on Token Plan: 1M context, up to ~393k output tokens; no benchmarks or pricing yet

Why it matters

【Key points】

· On 2026-11-11 at 10:01 Beijing time, Qwen announced "Qwen4 is coming" and opened a preview of Qwen4-Max (qwen4-max-preview), the first Qwen4 model.

· Pitched at complex coding, knowledge work, front-end builds and professional software workflows.

· Alibaba Cloud Model Studio docs list a 1M-token context, same as Qwen3.8-Max.

· Max output: 393,216 tokens (thinking + answer), about 3x Qwen3.8-Max's 128k.

· Text, image and video input: up to 16M pixels per image, 2-hour / 2 GB videos, 2,048 images or 64 videos per request.

· Hybrid thinking (on by default), function calling, built-in tools, structured output.

· API access is Token Plan only; also on QwenWork, Qoder, the Qwen app and QwenStudio.

· Not published: benchmarks, per-token pricing, parameters, open weights, GA date.

【Our read】

• · Background: a teaser plus preview, not a full launch.
• No blog, model card or official benchmark yet.
• Qwen has kept Max closed while opening large same-generation weights (Qwen3.8-2.4T-A95B in August); no word on Qwen4.

• · Comparison: 128k → ~393k max output targets long reasoning and large one-shot code.
• Kimi K3.1 Pro and DeepSeek V4.1-Flash-1017 shipped with open weights and scores; Fable/Mythos 5.2 and Gemini 4 Carbon are GA with evals and prices.
• Qwen4-Max can only be compared on specs for now.

• · Impact: pay-as-you-go API users can't call it yet.
• Launching via subscription and Alibaba's own apps shows subscriptions becoming a primary launch channel.
• Watch for the tech report, pricing, weights and GA.

【Verifiable】

• ·
• ·
• ·