The announcement and the claimed capabilities
Alibaba has announced the imminent release of the new generation of its Qwen language model, called Qwen4, according to IT之家. The company's communication speaks of deeper knowledge understanding, more complete world perception, and further development of coding and agentic capabilities.
For now, these are company statements rather than technical results that can be independently verified: the available material contains no benchmarks, technical specifications, model size, or details about the architecture. It is therefore impossible to determine how much the claimed capabilities represent a leap over previous versions.
The one firm point is that the new model family is presented as a generational update, not as a simple incremental revision. The announcement concerns both the model and the ecosystem of products that will use it.
The preview, pricing and products involved
According to the source, a preview version, Qwen4-Max, has already been made available through the 阿里 Token Plan service, with a promotional offer that allows users to try the model at a discounted price as low as one tenth of the usual cost. The immediate integration of the new model into several products is also indicated: 千问办公, Qoder, the 千问 app, and QwenStudio.
On the commercial side, IT之家 reports that the individual Token Plan, in the Lite and Essential variants, has a promotional price starting at 39 yuan per month; during the initiative, new customers can get a further 50% discount, bringing it to 19.5 yuan. These are stated promotional conditions, whose duration is not specified in the material.
The coexistence of an already active preview and an announcement of an upcoming release suggests a gradual distribution strategy: the model is made accessible before the official launch, while internal products are updated in parallel. However, the material does not clarify when the final release will take place or whether the preview differs from the final version.
What remains to be verified
The available information does not allow an assessment of Qwen4's actual performance or a comparison with other models. There is a lack of data on context, cost per token, geographic availability, licenses, and access methods for external developers. It is also not indicated whether and when the announced integrations will be extended to other services.
In summary, the announcement confirms the existence of a new cycle of Qwen models and the start of a paid early-access phase. Claims about capabilities should be treated as company statements pending independent verification and technical details.
