Alibaba Open-Sources Qwen3.8-Max, a 2.4 Trillion Parameter Model
2 min readAlibaba has published the open weights for Qwen3.8-Max, a 2.4 trillion parameter model, making it the first system at Qwen’s top Max tier that anyone can download and run. The weights went up on Hugging Face and ModelScope roughly a week after the model launched on QwenCloud.
What Qwen3.8-Max Actually Is
The model is a mixture of experts design built on the Qwen3.5 architecture: 2.4 trillion total parameters with about 95 billion active on any given query. It accepts text, image and video input and carries a one million token context window. Alibaba priced the hosted API at $2 per million input tokens and $6 per million output tokens, which is parity with OpenAI’s GPT-5.6 rather than the deep discount Chinese labs usually lead with.
The Pitch Is Endurance, Not Benchmarks
Alibaba’s launch material leans on long running autonomous work rather than single shot test scores. In one company run trial, reported by TestingCatalog, the model spent about 16 days building and maintaining an open source command line project, producing 265 commits, 127 pull requests and 151 issues through a loop of implementing, testing and repairing its own work. In a hardware design test it shrank a cryptographic accelerator from 8,298 gates to 678 across roughly 500 turns and reached timing closure at 500 MHz.
These are vendor numbers and have not been independently reproduced. The shape of the claim is still worth noting: the pitch is stamina measured in days, not cleverness measured in seconds.
Why an Open 2.4 Trillion Parameter Model Matters
Publishing weights at this scale changes who can inspect and self host a frontier class system. Regulated industries, academic groups and any organization that cannot ship data to a vendor endpoint now have an option that was previously closed to them. It also lands in a market already tilting: Chinese open weight models accounted for roughly 61% of tokens consumed on the OpenRouter aggregator by May 2026.
The strategy reads clearly enough. Matching American pricing on the hosted API while giving the weights away squeezes closed model economics from both ends at once. The question worth watching is whether Western enterprises actually deploy a 2.4 trillion parameter Chinese model in production, or whether procurement policy keeps it at arm’s length no matter what the benchmarks say.