Alibaba stock reacts to flagship AI preview
Alibaba's shares rallied on July 20 after the company unveiled Qwen3.8 Max-Preview, helping the Hang Seng gain ground as investors weighed the AI announcement. The Hong Kong-listed stock opened roughly 3 percent higher and the New York-listed ADRs added about 3 percent in pre-market trading, with the move framed by coverage that positioned the release as a direct counter to rival Chinese frontier models.
Alibaba positions the model against Claude Fable 5
Alibaba said on its official Qwen account on X on July 19 that Qwen3.8-Max-Preview, built with 2.4 trillion parameters, is "one of the most powerful models available today, comparable to leading frontier AI models, second only to Fable 5," referring to Anthropic's most widely released Claude system. The post was made during the World AI Conference in Shanghai, the country's largest annual AI summit, and underscored how Chinese labs now benchmark themselves against a single US frontier model rather than a broader field.
A direct response to Moonshot's Kimi K3
The preview arrives days after Moonshot AI, also backed by Alibaba, released Kimi K3, a 2.8-trillion-parameter model that pushed Fable 5 into second place on a closely watched AI coding leaderboard and jolted global technology stocks. Bloomberg reported that Moonshot is preparing a Hong Kong IPO within roughly six months at a valuation near 30 billion US dollars after the model crossed 300 million US dollars in annual recurring revenue in June, and that Qwen3.8 is widely seen as an attempt to blunt that momentum. One outlet quoted an analyst view that the release, coming only about two months after Qwen 3.7 in May, was timed to check Kimi K3's rapid ascent.
Availability, pricing, and open-weight plans
Developers can test the preview through Alibaba's Qoder and QoderWork coding platforms as well as a Token Plan tier for international and domestic users, with the coding platforms priced at about 10 percent of the standard cost. Alibaba said full model weights would be opened "soon" without committing to a specific date, a strategy that has helped Qwen gain ground on closed rivals such as Anthropic and OpenAI. The release also follows China's Cyberspace Administration approving Apple Intelligence for local release this month, a separate win for Alibaba's Qwen as a regional technology partner.
Missing benchmarks and analyst caution
Alibaba disclosed no independent benchmark scores, training-data size, activated-parameter count, or licensing terms alongside the preview, drawing scrutiny over how the 2.4 trillion parameter figure translates into real-world inference cost. Analyst commentary carried in Chinese-language coverage noted that earlier Qwen releases used a Mixture-of-Experts architecture in which only a slice of weights is active per token, meaning nominal parameters can overstate actual compute demand. Whether Qwen3.8 Max delivers on its Fable 5-comparable claims is likely to hinge on forthcoming benchmark results and the eventual open-weight release.
Follow-up signals
Alibaba has not given a precise date for releasing Qwen3.8 Max's open weights, and benchmark scores remain pending. The next verifiable milestones are the eventual public weight drop, independent third-party evaluations, and the next Kimi K3 access updates as Moonshot moves toward its reported Hong Kong listing within six months.
Share this article
