Alibaba (09988) launches open-source Qwen 3.6-35B-A3B model in a thousand questions platform

On April 16th, Alibaba (09988) officially released the open source of the Qwen3.6-35B-A3B model in the Qwen 3.6 series of medium-sized models. This model only activates 3B, not only lightweight and efficient, but also performs outstandingly in intelligent agent programming, significantly surpassing the previous generation model Qwen3.5-35B-A3B, and can compete with dense models such as Qwen3.5-27B and Gemma4-31B. The new model also supports multimodal thinking and non-thinking modes, making it one of the most versatile open source models currently available. According to information, Qwen3.6-35B-A3B is one of the representatives of the current global AI community's "efficient lightweight" open source models. It uses the Mixed Expert (MoE) architecture, with a total parameter volume of 35 billion and only 3 billion activations. In the inference process, it can achieve higher performance intelligent output with lower computing power consumption. In particular, it excels in intelligent agent programming tasks: in authoritative benchmark tests such as Terminal-Bench2.0 for terminal programming, long-range programming task NL2Repo, and real-world Agent capability evaluation QwenClawBench, the performance of Qwen3.6-35B-A3B is significantly better than the previous generation Qwen3.5-35B-A3B, as well as similar open source models like Gemma4-26B-A4B and Gemma4-31B. Qwen3.6-35B-A3B also achieves deep compatibility with mainstream agent frameworks such as OpenClaw, Qwen Code, and Claude Code, allowing the model's programming capabilities and native multimodal capabilities to be better endowed to various intelligent agents to complete longer and more complex tasks.