Qwen Team Open-Sources Qwen3.6-35B-A3B: A Sparse MoE Vision-Language Model with 3B Active Parameters and Agentic Coding Capabilities

Alibaba’s Qwen team released Qwen3.6-35B-A3B, an open-weight vision-language model.

It has 35B total parameters but only 3B active at inference, using a sparse Mixture-of-Experts design.

The model posts strong results on coding, reasoning, and multimodal benchmarks, including agentic coding and vision tasks.

The release highlights how sparse MoE models can improve efficiency and lower inference cost for open AI systems.