Summary: On November 11, 2026, the Qwen team announced on X that Qwen4 is coming, pointing to deeper knowledge, richer world understanding, and stronger coding and agentic capabilities — and confirming that Qwen4-Max is already in preview on Alibaba’s Token. Alibaba Cloud’s Model Studio docs now list the model as qwen4-max-preview, so builders can start probing it today, though only on the Token Plan.
What you can actually call right now
The docs position qwen4-max-preview as the quality-first option for “tasks with higher reasoning and understanding requirements,” while suggesting qwen3.7-plus as the capability-cost balance and qwen3.8-flash when cost comes first. The spec sheet is where the preview gets concrete: 1M context and a 393,216 max output budget that covers both thinking and answer tokens — notably higher than qwen3.8-max and qwen3.8-flash, which cap at 128k output on the same table.
Input handling matches the current flagship tier: text, images, and video go in, text comes out. Per-image pixel limit is 16M, video runs up to 2 hours or 2 GB per clip, with 2,048 images and 64 videos accepted. On features, the preview supports thinking mode, function calling, built-in tools (web search, code interpreter, web scraping), structured output, and batch inference — the same feature set Alibaba lists for its other current flagships.
If you’ve been tracking this line, Qwen3.8-Max went GA with 1M context and open weights earlier in this cycle; the preview continues that context-window story, at least on the API side.
What the announcement does not settle
The preview spec is not the full Qwen4 story. Neither source gives a general release date, benchmark numbers, or pricing beyond the Token Plan gate, and neither explains what specifically separates Qwen4-Max from qwen3.8-max beyond the positioning language and the larger output budget. The announcement post is also truncated, so the full framing isn’t visible.
Practical next step: run your hardest reasoning tasks against qwen4-max-preview and the same tasks against qwen3.8-flash — Alibaba’s own docs advise verifying the cheaper model first — and diff the results before committing budget to the Token Plan.
Sources
- Qwen on X: Qwen4 is coming. Deeper knowledge. Richer world understanding. Stronger coding a
- Text Generation and Visual Understanding - Alibaba Cloud
View the original post on X