Alibaba has opened a preview of Qwen4-Max, the first model from its next flagship Qwen4 family, pitching the system at long coding jobs and agent workflows that make repeated tool calls. Access is live now through the company's Token Plan subscription, with early access rolling out across its QwenWork, Qoder, Qwen App and QwenStudio products.
The company says Qwen4-Max improves on its predecessor, Qwen3.8, in world knowledge, code generation and sustained performance on long tasks. But Alibaba has published no benchmarks and no model card for the new system, so those claims are unverified for now — and the preview's headline numbers come from documentation and client code, not a full specification sheet.
What developers actually get
According to Alibaba's documentation, Qwen4-Max offers a 1M-token context window and up to 393,216 output tokens, with image and video input supported. The preview is available only to Token Plan subscribers for now — there is no pay-as-you-go route — with plans starting at about $3 a month for new users during a limited-time promotion that cuts the rate by 80 percent. Higher tiers add monthly credits and allow between one and eight agents running at once.
The service works with clients that speak OpenAI- or Anthropic-compatible protocols, including Qwen Code, Cline, Claude Code, Cursor, OpenCode and Codex, so developers can try the model by changing a base URL and API key rather than rebuilding their setup. Alongside the model, Alibaba shipped Qwen Code 1.0, which absorbs its Qoder coding tool into a single agent runtime.
A long tease, finally a launch
The launch caps weeks of signals. At the Apsara Conference in Amsterdam on October 29, a presentation listed Qwen4-Max, Qwen4-Flash, Qwen4-Plus and Qwen4-27B as forthcoming models, and Alibaba Cloud chief technology officer Feifei Li reportedly told attendees the family would arrive the following week. A stealth model codenamed Qoduck later surfaced inside Qoder, and Chinese outlet 53AI identified it as an early Qwen4-Max checkpoint — something Alibaba has not confirmed.
Paktechstan covered the first public tease on November 7, when Alibaba posted a cryptic "The Becoming Begins" message with no date and no specifications. The preview now answers the timing question, but leaves most of the technical ones open.
What is still missing
Alibaba has not published a parameter count, benchmark scores or a model card, and has said nothing about downloadable weights or a self-hosting licence. For a baseline, the predecessor Qwen3.8-Max reported scores of 86.6 on Terminal Bench 2.1, 67.7 on SWE-bench Pro and 93.0 on PaperBench — but those are the old model's numbers, not evidence about Qwen4-Max.
The launch also lands a week after Apple confirmed that Alibaba's Qwen will power Apple Intelligence in China, with testing that started on November 4 — a reminder that Qwen is now carrying one of the biggest consumer AI deployments in the world. What matters next is whether Alibaba publishes benchmarks, and whether the promised Plus, Flash and 27B tiers follow the flagship out of preview.



