Qwen 3.5-35B-A3B is beyond expectations. It's replaced GPT-OSS-120B as my daily driver and it's 1/3 the size.

Alibaba's Qwen research team has quietly released a model that's causing significant buzz in developer communities. The Qwen 3.5-35B-A3B, a 35 billion parameter model, is reportedly replacing much larger 120B parameter models as developers' daily driver for complex workflows. Early adopters describe using it for everything from code analysis and system generation to agentic tasks involving custom MCP (Model Context Protocol) servers and browser automation. What's particularly striking is that users report this smaller model handling their broad development needs—including priority-based message aggregation, timed task systems, and visual interpretation—while running at just one-third the size of competitors.

The technical details reveal why this matters: users are running the model quantized to Q4-K-XL format on consumer hardware like RTX 5090 and 3090 GPUs at 100,000 token context lengths. Its standout feature appears to be exceptional tool-use capability—knowing when it lacks information and effectively using browser access to fill knowledge gaps. While not the absolute smartest model available, its combination of reasonable intelligence, self-awareness of limitations, and efficient tool utilization makes it remarkably practical for production workflows. This development suggests we're entering an era where model efficiency and specialized capabilities might matter more than raw parameter count for many real-world applications.