FrontierThe story, in brief
Alibaba's open model Qwen3.6 leads Google's Gemma 4 across agentic coding benchmarks
Alibaba's open model just beat Google on coding benchmarks. With 35B parameters. Using only 3 at a time.
The KeyNews take
Why it matters
Alibaba's Qwen3.6 demonstrates that efficient sparse activation (MoE-style parameter sparsity) can outperform larger dense models on agentic coding tasks, challenging Google's approach and validating open-source competitive viability against closed alternatives.
The key facts
5 to knowQwen3.6-35B-A3B activates only 3 of 35B parameters per inference
Beats Google Gemma 4-31B on coding and reasoning benchmarks
Open-source model release
Sparse activation/MoE architecture advantage demonstrated
Agentic coding task performance as evaluation metric
Go to the source
The Decoderthe-decoder.com
Publisher excerpt: Alibaba's new open-source Qwen3.6-35B-A3B activates just three of its 35 billion parameters at a time, yet beats Google's larger Gemma 4-31B on coding and reasoning benchmarks.