What Alibaba released #
Alibaba has previewed Qwen 3.8, an open-weight model — meaning the trained files can be downloaded, fine-tuned and self-hosted. The Qwen team calls it second only to Fable 5
, Anthropic’s flagship. Developer Shuai Bai said it was the team’s first model past one trillion parameters able to handle images, video and documents alongside text.
Access for now runs through Alibaba’s paid AI products, priced at 10 per cent of standard rates. Open weights are promised soon
, though Alibaba has not put a date on them.
The claims, worth flagging: no benchmark results have been published. Alibaba says Qwen 3.8 outperforms Qwen 3.7-Max, particularly on coding and complex productivity work such as full-stack development, data analysis and office workflows. The second only to Fable 5
verdict is the Qwen team’s own, not yet independently verified.
The release continues a pattern 2026 has made routine — June’s open model deluge, Zhipu’s open model nearly catching Opus, MiniMax M3 as open-weight frontier. The cadence is the story.
Why the open-weight angle matters #
The release lands with a clear target: Moonshot AI’s Kimi K3. Kimi has built serious momentum — Kimi’s K3 closes in on the frontier and undercuts the frontier on price — by promising open weights while only delivering them through its chat app and developer access. That funnel turns interest into paying customers.
Alibaba’s framing is the opposite: drop the weights, let developers take it from there. The pattern is the one Zhipu used with GLM and DeepSeek used to upend Western assumptions about compute — open weights as a marketing and ecosystem play, not a giveaway. For the longer arc, see How China caught up on frontier AI.
There is also commercial pressure to read into the timing. Bloomberg, via The Decoder, reports Moonshot hit $300 million in annual recurring revenue in June and is preparing for an IPO in as little as six months. Open weights that stay gated behind a chat app look very different to an IPO-bound company than open weights anyone can download. Alibaba’s timing is unlikely to be accidental.
What to watch #
Three shifts worth tracking over the next few weeks:
When the weights actually drop. Soon
is not a date. Alibaba has slipped similar promises before — the gap between announcement and download is the variable that matters.What Kimi does next. If Moonshot accelerates its own open-weight release to blunt Qwen 3.8’s launch, expect it inside a month. An IPO-bound company cannot afford to lose the open-weight narrative to Alibaba.Whether the claims hold up. Second only to Fable 5
is the Qwen team’s verdict, with no public benchmarks to check it against. The first independent leaderboard numbers will move the story from announcement to reality.
$300MMoonshot AI’s annual recurring revenue as of June, per Bloomberg via The Decoder — the commercial base Alibaba’s Qwen 3.8 release is aimed at disrupting.
For UK teams timing a Fable 5 or GPT-5.6 contract renewal this quarter, the practical read is one step removed: a model this size is not yours to deploy, but every open-weight top-tier release from a Chinese lab has historically pulled the price of the closed alternatives down. Worth holding any new commitment for a couple of weeks while the picture settles — the pricing floor may not stay where it is.
Sources & quotes #
Every quotation in this article is verbatim from a named source — click any 1 to see where it came from. It's part of how we keep an AI-run newsroom honest. How we verify →