Alibaba has introduced its new flagship model, Qwen3.8, featuring a 2.4 trillion parameter architecture. Developers claim that the new model is capable of competing with Anthropic Fable 5, and an open-weight version is expected to be released in the near future.
What Happened
Alibaba announced Qwen3.8 with a 2.4 trillion parameter architecture. Currently, a preview version, Qwen3.8-Max-Preview, is available via the Token Plan subscription, as well as through the Qoder and QoderWork services. A full release of the open-weight model is planned for the near future.
Context
The scale of the model places it alongside the largest proprietary frontier models. The claim of comparability with Anthropic Fable 5 indicates Alibaba's ambition to secure a dominant position in the top-tier LLM segment.
Why It Matters for the Industry
The release of a model of this scale in an open-weight format could significantly narrow the gap between proprietary frontier models and open solutions. This will create a powerful catalyst for the development of complex agentic systems and stimulate the emergence of new optimization techniques, such as quantization and specialized serving via vLLM or TensorRT-LLM.
Why It Matters for Users
Users can begin testing the capabilities of Qwen3.8 now via API (Token Plan), integrating the model into their workflows using tools like Cursor or Claude Code. This allows for the evaluation of generation quality and reasoning before the official release of the weights.
What Is Currently Unknown / Limitations
There are concerns regarding the infrastructure costs and the complexity of local deployment (inference) of a model of this scale for most companies.
Sources
Author
Look at AI, Editorial Team