Qwen 4 Max Preview
A preview version of Alibaba's flagship large model. It supports multimodal understanding and tool calling, has a context window of one million tokens, and uses a Mixture-of-Experts (MoE) architecture.
Core Facts
The Qwen4 Max Preview runs until November 25.
Qwen2.5-Max-0902 supports a context window of 1 million tokens.
The Qwen repository merged an update that raises the default context for models beginning with Qwen4 from 256K to 1M Tokens, covering variants such as Max, Plus, and Coder, and names such as Qwen4 Max and Qwen4.5 Max appear in the test code.
Timeline (last 90 days)
Making long-term production decisions based on incomplete benchmarks and vague pricing is a gamble, not an engineering choice.
Qwen 4 Max Preview is paid for via subscription and credit multipliers, with Coder's launch promotion reducing the credit cost by 80%, and there is no clear per-token pricing.
Four days before launch, the coding tool Coder offered a mysterious model named CoDuck, and on launch day Coder confirmed that CoDuck was a test build of Qwen, updated daily, and has now been upgraded to Qwen 4 Max Preview.
In the ModelDial test, Qwen 4 Max Preview scored 90 on knowledge work, about 80 on back-end and front-end code, and only 59 on aesthetics and maintainability.
Alibaba released Qwen 4 Max Preview on November 11.
Qwen 4 Max Preview, the next-generation flagship model of Alibaba's Tongyi Qianwen family, has opened for preview, officially positioned as a multimodal flagship model for coding, office work, front-end design and complex agent tasks, supporting text, image and video input.
Qwen 4 Max Preview's token packages are restricted to interactive use within tools such as Claude Code, Cursor and Qwen Code, and are explicitly prohibited for automated scripts or batch tasks, with violations potentially leading to key bans; personal packages do not support cancellation once activated.
The most widely circulated pelican-riding failure case on X (about 195,000 views) had no prompt disclosed by the original author, and the creator judged that the viral case was most likely a single stroke of bad luck or the use of a prompt that would not be made public, rather than a systematic flaw of the model.
In a bug-hunting test of about 3,000 lines across 42 files with 16 real bugs and 11 decoys planted, Opus 5.5 fixed 15/16 with zero regressions in just over 8 minutes for $2.24; Qwen 4 Max Preview had no score because the chat app could not open the project folder and the ultra-long input was blocked by a human verification check.
In the pelican-riding-a-bicycle test, Qwen 4 Max Preview completed a four-camera 3D animation in about 24 minutes on the second attempt without the wheel-flipping problem seen in the viral clip; Opus 5.5 took 22 minutes at about $5.40, additionally including a basket, seagulls, rocks and one-click switching among four viewpoints.
28 more timeline events
All Facts (20)
The Qwen4 Max Preview runs until November 25.
70%VerifiedQwen2.5-Max-0902 supports a context window of 1 million tokens.
70%VerifiedThe Qwen repository merged an update that raises the default context for models beginning with Qwen4 from 256K to 1M Tokens, covering variants such as Max, Plus, and Coder, and names such as Qwen4 Max and Qwen4.5 Max appear in the test code.
65%Partially VerifiedCurrently Qwen 4 Max is available only in preview form, with no open weights, and model files cannot be downloaded.
65%Partially VerifiedAs a preview version, Qwen 4 Max's performance fluctuates considerably, so it is recommended to try it first on non-critical paths and do a full evaluation after the official version is released.
65%Partially VerifiedThe original material does not give Qwen4-Max's specific benchmark scores, evaluation set names, or list of comparison models.
65%Partially VerifiedQwen4-Max is positioned as a reasoning model, natively supports visual input, and has a context window of 1 million tokens.
65%Partially VerifiedQwen4-Max actually runs what it builds and corrects camera, animation and lighting issues through visual feedback.
65%UnverifiedThe Qwen 4 Max preview ends on November 25, after which the model will be retired.
60%UnverifiedThe Qwen 4 Max preview cannot be downloaded and has no token-billed API; the only entry point is Alibaba Cloud's token package, billed by monthly subscription.
60%UnverifiedIndependent evaluation organization Model Dial tested Qwen 4 Max Preview, which scored 78.8/100, ranking 14th among 54 models.
60%UnverifiedIn the Atomic Agent team test, Qwen 4 Max passed all 59 hidden tests in three out of three runs, while the previous generation Qwen 3.8 Max failed to complete even once; Qwen 4 Max took about 25 minutes, Opus 5.5 about 4 minutes, and GPT-6.1 Sol less than two and a half minutes, and Sol cost about 29 cents while Qwen cost about 65 cents.
60%UnverifiedQwen 4 Max Preview is paid for via subscription and credit multipliers, with Coder's launch promotion reducing the credit cost by 80%, and there is no clear per-token pricing.
50%UnverifiedIn the ModelDial test, Qwen 4 Max Preview scored 90 on knowledge work, about 80 on back-end and front-end code, and only 59 on aesthetics and maintainability.
50%UnverifiedMaking long-term production decisions based on incomplete benchmarks and vague pricing is a gamble, not an engineering choice.
50%UnverifiedThe most widely circulated pelican-riding failure case on X (about 195,000 views) had no prompt disclosed by the original author, and the creator judged that the viral case was most likely a single stroke of bad luck or the use of a prompt that would not be made public, rather than a systematic flaw of the model.
50%UnverifiedQwen 4 Max Preview's token packages are restricted to interactive use within tools such as Claude Code, Cursor and Qwen Code, and are explicitly prohibited for automated scripts or batch tasks, with violations potentially leading to key bans; personal packages do not support cancellation once activated.
50%UnverifiedDeveloper XODUD confirmed that Qwen 4 Max Preview runs only under the token package model before November 25, with no pay-as-you-go option.
50%UnverifiedQwen 4 Max Preview offers discounts of up to 80% during the promotion, with an equivalent input cost of about 2-3 USD per million tokens under the token package model.
50%UnverifiedOn the GameFeel RE benchmark (covering more than 8,000 programming and game design tasks), Qwen 4 Max Preview scored 10.3.
50%