Alibaba released Qwen 3.8, a multimodal AI model with 2.4 trillion parameters, claiming it rivals leading models and is second only to Fable 5. A preview is now available.
SynthePulse Insight · AI deep reading
Version 1 · 1 source
Alibaba's Qwen team announces the upcoming release of the open-source Qwen 3.8 model with 2.4 trillion parameters, claiming performance second only to Anthropic's Claude Fable 5. This release marks another leap in parameter scale for Chinese open-source large models.
On July 19, 2026, Alibaba's Qwen team announced on X platform that the Qwen 3.8 model will be released soon with open weights. According to a tweet from the official account @Alibaba_Qwen, the model has 2.4 trillion parameters, making it one of the largest open-source models by parameter count, second only to Anthropic's Claude Fable 5.
The news was reposted by X user meng shao, whose tweet stated 'Chinese open-source models are sprinting' and cited the Qwen official account's statement. Currently, this news has not appeared in Alibaba's official press releases or technical blogs, so it is in the product preview stage.
The Qwen team claims that Qwen 3.8 is 'one of the most powerful models currently available, comparable to leading frontier AI models, second only to Fable 5.' This statement directly positions Qwen 3.8 against Anthropic's flagship model Claude Fable 5, which is believed to have a larger parameter count (specifics not disclosed), but Qwen 3.8 holds a leading position in parameter scale among open-source models.
It should be noted that these performance claims are entirely based on Alibaba's own statements, lacking third-party benchmark tests or independent verification. In the open-source community, actual model performance typically requires confirmation through third-party evaluations (e.g., LMSYS Chatbot Arena, MMLU, HumanEval, etc.).
Qwen 3.8 will be released as 'open-weight,' meaning the model weights will be made public, but training code, datasets, and detailed technical reports may not be released simultaneously. This is consistent with Alibaba's previous open-source strategy for the Qwen series.
Currently, there are several uncertainties: the specific release date is not set; the open-source license (e.g., whether commercial use is allowed) has not been disclosed; the actual inference cost and deployment requirements (2.4T parameters require extremely high compute) are also unspecified. Additionally, parameter count does not linearly correlate with performance; factors such as model architecture and training data quality are equally critical.
This report is primarily based on an official account statement on X platform and a user repost, and is in the product preview stage. All performance claims come from Alibaba itself and have not been verified by third parties. The parameter count of 2.4T is an official figure, but the ranking of 'second only to Fable 5' lacks public benchmark support.
Qwen 3.8 sets a new benchmark in parameter scale for open-source models with 2.4T parameters, but actual performance still requires independent evaluation. Its open-source release will intensify competition in the large model space, especially between Chinese and American open-source models.
Primary report
Primary source
Also covered by 1 sources