阿里巴巴开源模型 Qwen3.6 在智能体编码基准测试中领先谷歌 Gemma 4

Alibaba's open model Qwen3.6 leads Google's Gemma 4 across agentic coding benchmarks

AI 导读

阿里巴巴发布新一代开源模型 Qwen3.6-35B-A3B，采用稀疏激活架构，总参数量达 350 亿但推理时仅激活 30 亿参数。该模型在多项编码与推理基准测试中击败谷歌 310 亿参数规模的 Gemma 4-31B，以更小激活参数量实现更强智能体编码能力，展现高效架构优势。

正文 · 原文

Alibaba has released Qwen3.6-35B-A3B, a new open AI model. The mixture-of-experts model activates just three of its 35 billion parameters at a time, cutting compute costs without meaningfully hurting quality, according to Alibaba.

Alibaba says the model significantly outperforms its predecessor, Qwen3.5-35B-A3B, on agentic coding tasks. Against Google's open Gemma 4-31B, it leads every coding benchmark listed, scoring 73.4 to 52.0 on SWE-bench Verified and 51.5 to 42.9 on Terminal-Bench 2.0. It also edges ahead on reasoning tests like GPQA (86.0 to 84.3) and AIME26 (92.7 to 89.2). Alibaba claims it even keeps pace with Claude Sonnet 4.5 on image and video tasks.

Benchmark results show Qwen3.6-35B-A3B leading across coding, reasoning, and multimodal tests against Qwen3.5 and Google's Gemma 4 models. | Image: Alibaba / Qwen

The model offers both thinking and non-thinking modes. Users can try it in Qwen Studio, access it via API as Qwen3.6 Flash through Alibaba Cloud Model Studio, or download the weights from Hugging Face and ModelScope. The release follows the launch of the larger Qwen3.6-Plus.