Alibaba's Qwen3.6 Outperforms Google's Gemma 4 on Coding Benchmarks

Alibaba's new open-source AI model Qwen3.6-35B-A3B, with only 3 of its 35 billion parameters activated, outperforms Google's larger Gemma 4-31B on coding and reasoning benchmarks.

Details

Alibaba has developed a new open-source AI model called Qwen3.6-35B-A3B that is capable of outperforming Google's larger Gemma 4-31B model on coding and reasoning benchmarks. The key innovation in Qwen3.6 is that it only activates 3 of its 35 billion parameters at a time, allowing for highly efficient and targeted inference. This sparse activation approach enables Qwen3.6 to match or exceed the performance of larger, more computationally intensive models like Gemma 4. The results demonstrate the potential for more compact, energy-efficient AI models to achieve state-of-the-art capabilities across a range of tasks.