Alibaba Launches Qwen4-Max-Preview, Access from November 11 to 25

Key Takeaways

1Access announced from November 11 to November 25, 2026, reserved for Token Plan subscribers

2Multimodal model with Function Calling, OpenAI/Anthropic APIs, and 1,000,000 tokens of context

3First third-party scores: 10.3 on GameFeel RE and 58.8/100 on StealthMark

Why it matters : the trial window is short and official evaluations are lacking, while independent tests provide partial benchmarks for a model still in pre-release.

Alibaba offers Qwen4-Max-Preview for a short trial window reserved for subscribers. The model is multimodal, compatible with APIs from OpenAI and Anthropic, and features a very large context memory. Initial third-party tests deliver accurate but partial scores, while no official comparison benchmark has been published.

Third-party scores exist, but no official comparison table

Alibaba has not published any comparative benchmark tables for Qwen4-Max-Preview. However, independent measurements provide some reference points. On GameFeel RE, released by Offloop on November 11, 2026, Qwen4-Max-Preview scores 10.3 points, while Qwen3.8-Max reaches 12.1, Qwen3.8-Flash-Next 13, Qwen4-27B Early 14.9, and Qwen4-Flash Early 19.9. In the StealthMark test by Stealth Models, the model scores 58.8 out of 100 and ranks 12th out of 58 evaluated models, with notable performance in certain visual creation tasks. Offloop also indicates that Qwen4-Max-Preview can generate games and interactive scenes, but Claude Opus 5.5 had a slight edge in a comparative game creation test. Further independent evaluations will be necessary to assess its overall performance.

Le brief IA que lisent les pros

Le brief IA que les pros lisent chaque soir

Les 7 actus IA du jour, décryptées en 5 min. Gratuit.

Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.

Choisissez votre rythme

Gratuit · Pas de spam · Désabonnement en 1 clic

Temporary access reserved for Token Plan subscribers

Access to Qwen4-Max-Preview is limited from November 11 to 25, 2026, with a planned withdrawal at the end of this period. Its use requires a subscription to the Token Plan, available in Personal or Team formats, as Pay-as-you-go pricing is not compatible. The model can be tried via QwenWork, Qoder, the Qwen app, and QwenStudio. Access conditions may vary by channel, and free access is not guaranteed.

Multimodality and tasks announced by Alibaba

Qwen4-Max-Preview is a multimodal model that supports the processing of images, screenshots, and videos. According to Alibaba, it is capable of solving complex problems, performing video analysis, and utilizing tools. The company also states that this model has been designed for code generation, web interface creation, and executing office tasks. Qwen4-Max-Preview offers a context window of one million tokens and can produce up to 393,216 tokens in output.

Tools, APIs, and integration into workflows

The model can be integrated into automated workflows driven by AI agents and supports Function Calling to use external tools. It accommodates various API formats, including those offered by OpenAI and Anthropic. Developers have the option to add it to existing applications without needing to modify their entire architecture.