Speculation Arises as Qwen Developer Advises Against Waiting for 35B-A3B Model
A Qwen developer reportedly advised the community not to wait for the release of the 35B-A3B model, sparking speculation about Alibaba's upcoming model roadmap. This has led users to wonder if a different model, such as a larger 122B version, is planned instead. The 35B-A3B model is highly anticipated as a Mixture-of-Experts (MoE) model that offers high performance with low active parameter requirements, making it ideal for local deployment. If this model is delayed or bypassed, it could shift how developers plan their local AI hardware setups and application deployments. The Qwen 3.6-35B-A3B model features 35 billion total parameters but only activates 3 billion parameters per token, allowing it to run efficiently on consumer-grade GPUs. Speculation suggests that the developer's comment might point toward the release of a larger model or a shift to newer versions like Qwen 3.8.
## BACKGROUND
Qwen is a family of open-weight large language models developed by Alibaba Cloud. The "35B-A3B" designation refers to a Mixture of Experts (MoE) architecture, where the model has 35 billion total parameters but only activates 3 billion active parameters per token during inference, significantly reducing computational requirements and generation time.