Over the last month, the west has been stunned by the performance of GLM 5.2, then by Kimi K3, but it appears that Chinese labs aren’t close to being done yet.
Alibaba has now put its own flagship on the table, and the numbers on the page are hard to ignore. The company has unveiled Qwen3.8-Max-Preview, a 2.4 trillion-parameter model that Alibaba is calling one of the most capable systems available anywhere today, describing it as trailing only Claude Fable 5 among the models it benchmarked against. The model has gone live on Alibaba’s Token Plan, alongside the company’s coding-focused Qoder and QoderWork platforms, and Alibaba says open weights will follow.
Qwen3.8-Max-Preview arrives less than three months after Qwen 3.6 Max Preview, which itself had marked a departure from Alibaba’s usual playbook of releasing its top models as open weights from day one. That release stayed closed, hosted only through Qwen Studio and Alibaba Cloud. Qwen3.8 appears to be following a similar path for now, launching as a proprietary preview before an eventual open release, a pattern Alibaba has used before scaling up its biggest models for public download.
The parameter count alone marks a significant jump. Alibaba’s first trillion-parameter model, Qwen3-Max-Preview, launched back in September 2025 and was the company’s first model to cross that threshold. Qwen 3.6 Max Preview followed in April this year. Going from that to 2.4 trillion parameters within a matter of months puts Alibaba in the same scale bracket as Moonshot AI’s Kimi K3, which launched at 2.8 trillion parameters and briefly held the title of the largest open-source model ever released.
That comparison isn’t incidental. Alibaba owns roughly 36% of Moonshot AI, the Beijing-based lab behind Kimi, making Qwen3.8’s launch something of an in-house rivalry as much as a competitive one against Kimi K3 and Z.AI’s GLM 5.2. Alibaba has backed most of China’s prominent AI labs in some form, but the Moonshot stake is its largest single bet, and having two of its portfolio companies trade blows for the title of best open model in the world is not a bad problem for Alibaba to have.
The broader picture is one where Chinese labs are no longer treating open-weight releases as a consolation prize next to what OpenAI, Google, and Anthropic ship behind APIs. GLM 5.2 landed within striking distance of Claude Opus 4.8 on FrontierSWE and drew praise from executives at Vercel and Box. Kimi K3 went further, beating Claude Fable 5 and GPT-5.6 Sol outright on Arena’s Frontend Code leaderboard, becoming the first open-weight model to top that particular ranking. Qwen3.8 now joins that conversation with a parameter count larger than either.
Alibaba’s pitch around the release leans heavily on access rather than just capability. Qwen3.8-Max-Preview is being bundled into the company’s Token Plan subscriptions, which range from a $6 Lite tier offering 2,500 credits every seven days up to a $68 Pro tier with 40,000 credits over the same window and support for running six to eight agents concurrently. The plans also carry compatibility with tools built around OpenAI and Anthropic’s protocols, meaning developers using existing coding agents and harnesses can point them at Qwen3.8 without rebuilding their workflows.
What Alibaba hasn’t done yet is publish independent third-party benchmarks for Qwen3.8, so the “second only to Fable” claim currently rests on the company’s own internal evaluations, the same position Moonshot AI took when it first announced Kimi K3 as trailing only Fable 5 and GPT-5.6 Sol. Whether Qwen3.8 holds up once outlets like Artificial Analysis and LMArena run their own numbers will decide how much of this claim survives contact with independent testing.
For now, though, the trend is unmistakable. Three major Chinese releases in as many months, each claiming a spot at or near the very top of the global leaderboard, and each one open-weight or headed there. Whatever the final scores say, Alibaba, Moonshot, and Z.AI between them have made it very difficult to talk about frontier AI without talking about China.