Alibaba Claims Qwen3.8 Trails Only Anthropic’s Fable 5

Alibaba has previewed the most capable model in its Qwen line and says it ranks second among the world’s frontier systems, behind only Anthropic’s Claude Fable 5. The claim arrived without the one thing that would let anyone check it: no benchmark scores, no model card, and no independent evaluation.

The model, a preview build called Qwen3.8-Max-Preview, went live on the Qwen team’s account in a post on X during the World Artificial Intelligence Conference in Shanghai on July 19, 2026, described as “second only to Fable 5” and comparable to the leading frontier systems. It carries 2.4 trillion parameters and is the first Qwen model past the trillion-parameter mark to handle images, video and documents, not just text — a shift that moves Qwen from a text engine toward the kind of general-purpose system its US rivals already ship.

The preview is available now through Alibaba’s Token Plan subscription and its Qoder and QoderWork developer tools at 10% of the standard rate during the trial, with open weights promised but no release date or license terms disclosed.

A ranking with no scorecard

Parameter count describes a model’s size, not its skill. A 2.4-trillion-parameter figure says nothing on its own about how Qwen3.8 reasons, plans or holds up on real work — that is what benchmarks and evaluations are for, and it is precisely what Alibaba left out.

The company published no model card, no count of how many parameters are active on each query, and no task-level results, not even against its own previous flagship. It described Qwen3.8-Max only as “continuously evolving.” Each of those gaps matters to a buyer: an activated-parameter count signals how much compute a query burns, a model card documents training data and safety testing, and task-level scores show where a model is strong and where it is not.

That silence breaks from how Alibaba shipped its last top model. Qwen3.7-Max arrived in May 2026 with a full set of published numbers, including a score of 56.6 on the independent Artificial Analysis Intelligence Index. A vendor’s own benchmark table is a claim rather than an independent measurement, but it at least gives outside evaluators something to run down — the kind of scrutiny that has repeatedly upended assumptions about what these systems can actually do. The Fable 5 comparison offers nothing to check.

The independent picture that does exist points the other way. No outside leaderboard has ranked Qwen3.8 yet, and the most recent Qwen model that was scored, Qwen3.7-Max, sits far from the top of LMArena‘s list, the crowd-voted leaderboard where Fable 5 currently holds first place. Alibaba is asking buyers to accept a jump from the middle of the pack to second in the world on its word alone.

The China frontier race

The timing is not accidental. Moonshot AI unveiled its own 2.8-trillion-parameter model, Kimi K3, three days before Alibaba’s preview, moving China’s developers into the multi-trillion-parameter class that until recently belonged to the largest US labs. The rivalry is partly in-house — Alibaba is among Moonshot’s backers, which leaves two of its own bets trading blows for the title of China’s strongest open model. Both launches also lean on the open-weight strategy that US labs are now racing to match.

The contrast between the two releases sharpens the point about evidence. Kimi K3 shipped with third-party scrutiny: Artificial Analysis ranked it third on its intelligence index, comparable to Anthropic’s Claude Opus 4.8. That kind of outside check is exactly what Qwen3.8 lacks.

Chinese models have been gaining ground with Western developers partly because they cost less to run than the top American systems, and US lawmakers have begun weighing how to curb their adoption at home. For Alibaba, the stakes are competitive as much as technical. The company has spent the past year positioning Qwen as China’s default AI supplier, shipping a steady run of open-weight models and building its own cloud infrastructure and custom silicon to serve them — an effort that has drawn more developers to Qwen than to almost any other Chinese model family.

Bank of America (BAC ) analysts led by Alex Liu wrote that Kimi K3 has raised the capability ceiling for Chinese models and that Alibaba’s standing as the country’s open-source leader may face fresh tests. Investors were unbothered: Alibaba’s shares rose as much as 5.4% the day after the preview.

The ranking also traveled through friendly channels. The South China Morning Post, which Alibaba owns, relayed the Fable 5 comparison and the company’s description of Qwen3.8 as one of the most powerful models available today. Until Alibaba publishes benchmarks or the open weights it has promised, that ranking is a marketing line rather than a measured result — and for now the only people who can test it are the developers paying the discounted preview rate.