Alibaba’s Qwen 3.8 Max Claims to Rank Second Only to Claude Fable 5

Alibaba has introduced a preview of its latest AI model, Qwen 3.8 Max, during the World AI Conference in Shanghai. The company describes it as a 2.4 trillion-parameter model and says it ranks second only to Anthropic’s Claude Fable 5, based on its own internal testing.

The announcement came two days after Moonshot AI released Kimi K3, and some developers have been drawing parallels between the two. Kimi K3 was released as an open-weight model with 2.8 trillion parameters, making it for a short time the largest publicly available AI model.

Preview availability

The preview version is called Qwen3.8-Max-Preview and is available on Alibaba Cloud’s Token Plan subscription and coding platforms Qoder and QoderWork. It is available at a 10 percent discount from the standard price during the preview period.

Although Alibaba has yet to publish a formal specifications page, some technical details have been released via Alibaba’s integration documentation. According to that documentation, the model can handle a context window of nearly 984,000 tokens and a max output of 131,072. Reasoning effort can be set to low, high, or extra high (default).

Missing Technical Details

Alibaba has not released the benchmark results, model card, or licensing information for the preview model. The company has also not revealed the number of the model’s 2.4 trillion parameters that are currently used for inference, a detail that is used to estimate the computational cost of Mixture of Experts models.

According to the AI research site eesel AI, Qwen has gained a reputation for past releases that make performance claims without providing backup data. The outlet said that without a benchmark table, a claim of ranking second cannot be independently verified.

Alibaba’s history with large models

Qwen 3.8 is not Alibaba’s first model to leap the trillion-parameter line. In September 2025, Qwen3-Max-Preview hit that milestone, and then Qwen 3.6 Max Preview did in April 2026. The figure of 2.4 trillion parameters is a big step up from the previous model and is just within one year of Qwen 3.8.

There is also a notable ownership connection between the two companies involved. Alibaba holds an approximate 36 percent stake in Moonshot AI, the company behind Kimi K3, meaning both models come from firms in which Alibaba has a financial interest.

Industry Reaction

Some developers online suggested that the timing of the announcement of Qwen 3.8 was not accidental, given that it comes right after the release of Kimi K3. Whatever Alibaba did or did not intend to do, it’s evident that the trend this year is for a series of trillion-parameter, open-weight-bound models being released by multiple labs in China.

It’s also important to note that the Qwen 3.6 Max Preview was not released with open weights as Alibaba did for its flagship models. The version remained closed and available only through Qwen Studio and Alibaba Cloud. Qwen 3.8 appears to be following a similar approach so far.

What remains unclear

Alibaba has not shared a timeline, license, or benchmark data for open weights of Qwen 3.8 yet, but has claimed they will be published later. Until independent benchmark platforms evaluate the model, developers will need to rely on their own testing rather than Alibaba’s internal comparisons.

The next sign to watch will be whether outside evaluators such as LMArena or Artificial Analysis publish independent results once the model becomes more widely accessible.