aipopdb

22 Nov 2026  ·  10,137 works · 1,192 news indexed

News / Alibaba opens a 15-day preview of its flagship Qwen4-Max model

Alibaba opens a 15-day preview of its flagship Qwen4-Max model

AIPopDB Newsroom · via neurohive.io · · 3 min read · Comment

Read the original ↗
Image for Alibaba opens a 15-day preview of its flagship Qwen4-Max model

Image: Neurohive

Alibaba's Qwen team opened an early preview of Qwen4-Max on 11 November, the flagship of its next-generation Qwen family and the first model of the lineup to ship. The announcement was a short post headed "Qwen4 is coming" on X and Weibo. No technical report, model card or benchmark table came with it, only the post and two documentation pages.

Access runs through the paid Token Plan subscription and four Alibaba products: the Qwen app, Qoder, QwenWork and QwenStudio. The preview lasts 15 days and is switched off after 25 November, five days before the new-user price window closes. Plans use bundled Credits instead of per-token pricing, and during the preview the model draws as little as 20% of the standard Credit rate, falling to 8% between 22:00 and 08:00 Beijing time. The entry tier starts at about $3 a month for new users, a price that holds until 23:59 Beijing time on 30 November, and an extra 20,000-Credit bundle costs $15. With Credits instead of per-token prices, the real cost depends on the plan, the hour and the preview discount.

“Don't ask where it came from.”

Qoder, the line shown beside the stealth model Qoduck in its model picker on 7 Nov, in the Claude blog post · neurohive.io ↗

The model takes a 1-million-token context window and image and video input. It is compatible with the OpenAI Chat Completions and Responses APIs and with Anthropic's Messages API, so it works in tools such as Claude Code, Cursor, Cline, Qwen Code, OpenCode and Codex, and Qwen Code's default output limit is 32,768 tokens. It supports function calling, web search, built-in tools and JSON Schema output. One change for developers: max_tokens now caps reasoning and answer together, and the default reasoning level is xhigh. Essential, Standard and Pro plans allow one to eight concurrent agents.

Alibaba pitches it at real-world work: complex code, knowledge work, front-end builds and professional software, promising deeper knowledge, richer world understanding and stronger coding and agentic skills. AlphaSignal reports claimed gains over Qwen3.8 in world knowledge, code and long tool-calling tasks. Both analysis outlets stress that every quality claim so far comes from Alibaba. Parameter count, licence, open-weights plans and per-token price are unpublished. A chart showing a score of 53 on the Artificial Analysis index circulates, but the model is not in that site's rankings, and social posts claiming it beats GPT-6 and Claude Fable 5 are unsupported.

AlphaSignal's news card for the Qwen4-Max launch

AlphaSignal's news card for the Qwen4-Max launch

Image: AlphaSignal

Neurohive believes a stealth model called Qoduck was Qwen4-Max or an early build; Alibaba and Qoder have not confirmed that. Qoder added Qoduck on 7 November and removed it at 10:00 Singapore time on 11 November, the minute the preview went live. Early testers praised single-file builds such as a kanban board and an SVG pelican on a bicycle, while the Chinese outlet 53AI found it fast but uneven on multi-step tasks and weak at code completion for a Max-tier model.

Qwen4-Max
AI model
by Alibaba

About Qwen4-Max

Qwen4-Max is the flagship of the next generation of Qwen, the family of large language models from Alibaba Cloud, most of them released with open weights. Qwen began as a beta in April 2023 and opened to the public in September 2023, after regulatory clearance.

Alibaba has shipped Max-tier flagships before. It launched Qwen2.5-Max on 29 January 2025, and the previous flagship, Qwen3.8-Max, followed on 3 August 2026 with 2.4 trillion total parameters (95 billion active), a 1-million-token context and API access on Alibaba Cloud Model Studio. That model was previewed first, on 19 July 2026 on Token Plan and Qoder with a deep Credit discount, 15 days before its full release.

Nov 2026

Released

The pattern echoes Qwen3.8-Max, previewed on 19 July and released in full on 3 August, 15 days later. If Alibaba repeats it, the full Qwen4-Max could arrive in late November or early December, though that is Neurohive's estimate, not an Alibaba date. The Qwen4 lineup of Max, Flash, Plus and an open 27B was shown at an Apsara conference, with sources disagreeing on whether that was Hangzhou on 22 September or Amsterdam on 29 October; AlphaSignal's version has Alibaba Cloud CTO Feifei Li reportedly saying in Amsterdam that it would arrive the following week. Only Max has shipped, and no dates exist for Qwen4-Flash, Qwen4-Plus or Qwen4-27B.

Comments