Alibaba's biggest model ever is going open-weights

Alibaba Opens Up Its Biggest Model Yet

@AlibabaGroup has launched Qwen3.8-Max, its most capable model to date, and made a move that breaks with its own recent history. Qwen3.8-Max is the first model in the Qwen-Max class whose weights will be made publicly available. The move marks Alibaba's return to open-sourcing its top-tier AI models after keeping several recent flagship releases proprietary earlier this year. A second, smaller checkpoint called Qwen3.8-27B is also going open-weights alongside it.

Built on the Qwen 3.5 architecture, Qwen3.8-Max contains 2.4 trillion parameters, with its sparse mixture-of-experts design activating only 95 billion parameters during inference. The model supports a context window of up to 1 million tokens, meaning it can process up to 750,000 words or their equivalent for each query. API pricing comes in at $2.00 per million input tokens, $6.00 per million output tokens, and $0.25 per million cached tokens, undercutting many Western rivals by a significant margin.

On benchmarks, Alibaba claims Qwen3.8-Max ranks just below Anthropic's Fable 5, based on its own internal evaluations. The benchmark table, model card, and license have not been published. The original copy notes that Alibaba's own scorecard shows Fable 5 winning 15 of 31 benchmarks against Qwen's seven. No independent third-party benchmarks are yet available, which remains a reason for caution.

Chinese Models Are Winning the Distribution War

Whatever the benchmark results show, usage data tells a different story. Chinese-developed open-weight models held roughly 61 percent of token volume among OpenRouter's top 10 models in the week of February 24, 2026, a shift that happened inside eighteen months, with Chinese models going from under 2 percent of OpenRouter traffic in late 2024 to consistent majority leadership in 2026. Chinese AI models have since set a record 58 percent share of tokens processed by US firms on OpenRouter, briefly hitting 63 percent in early July, with DeepSeek the single largest vendor at 17.6 percent of routed tokens weekly and @Alibaba_Qwen next at 13.9 percent. The @Alibaba_Qwen series has also become the most self-hosted model in the world.

The practical caveat on the open-weights release is significant. A 2.4-trillion-parameter checkpoint is a multi-node datacentre artifact. No individual developer can run it on a workstation, and most companies cannot run it either. A distilled or quantized variant will likely determine how widely the release is actually adopted. At 4-bit precision, a 2.4-trillion-parameter model needs roughly 1.2 terabytes for weights alone.

Sources
South China Morning Post: Alibaba's AI model Qwen3.8-Max made widely accessible ahead of open-weights release
MarkTechPost: Alibaba Qwen Releases Qwen3.8-Max
The Decoder: Alibaba's open-weight Qwen3.8-Max takes on long-horizon AI tasks with 2.4 trillion parameters