English
Back
Open Account
智通财经APP
wrote a column · Apr 17 18:26

Alibaba (09988) Qwen 3.6-35B-A3B model open-sourced

According to Zhitong Finance APP, on April 16, Alibaba (09988) $BABA-W (09988.HK)$ The medium-sized model Qwen 3.6-35B-A3B in the Qwen 3.6 series has been officially open-sourced. This model only activates 3 billion parameters, making it not only lightweight and efficient but also excelling in agent programming, significantly surpassing the previous generation model Qwen 3.5-35B-A3B, and can compete with dense models like Qwen 3.5-27B and Gemma4-31B. The new model also supports multi-modal thinking and non-thinking modes, making it one of the most versatile open-source models currently available.
According to Zhitong Finance APP, on April 16, Alibaba (09988) $BABA-W (09988.HK)$ The medium-sized model Qwen 3.6-35B-A3B in the Qwen 3.6 series has been officially open-sourced. This model only activates 3 billion parameters, making it not only lightweight and efficient but also excelling in agent programming, significantly surpassing the previous generation model Qwen 3.5-35B-A3B, and can compete with dense models like Qwen 3.5-27B and Gemma4-31B. The new model also supports multi-modal thinking and non-thinking modes, making it one of the most versatile open-source models currently available. Data shows that Qwen 3.6-35B-A3B is one of the representatives of 'efficient lightweight' open-source models in the global AI community. It adopts a Mixture of Experts (MoE) architecture, with a total of 35 billion parameters and only 3 billion activated. During inference, it achieves high-performance intelligent output with lower computational power consumption. In particular, it performs exceptionally well in agent programming tasks: In authoritative benchmark tests such as Terminal-Bench2.0 for terminal programming, NL2Repo for long-term programming tasks, and QwenClawBench for evaluating real-world agent capabilities, Qwen 3.6-35B-A3B outperforms the previous generation Qwen 3.5-35B-A3B, as well as similar open-source models like Gemma4-26B-A4B and G...
Data shows that Qwen 3.6-35B-A3B is one of the representatives of 'efficient lightweight' open-source models in the global AI community. It adopts a Mixture of Experts (MoE) architecture, with a total of 35 billion parameters and only 3 billion activated. During inference, it achieves high-performance intelligent output with lower computational power consumption.
In particular, it performs exceptionally well in agent programming tasks: In authoritative benchmark tests such as Terminal-Bench2.0 for terminal programming, NL2Repo for long-term programming tasks, and QwenClawBench for evaluating real-world agent capabilities, Qwen 3.6-35B-A3B outperforms the previous generation Qwen 3.5-35B-A3B, as well as similar open-source models like Gemma4-26B-A4B and Gemma4-31B.
Qwen3.6-35B-A3B has also achieved deep compatibility with mainstream Agent frameworks such as OpenClaw, Qwen Code, and Claude Code, enabling the model's programming capabilities and native multimodal capabilities to be better integrated into various intelligent agents, allowing them to handle longer-term and more complex tasks.
Risk Disclaimer: The above content only represents the author's view. It does not represent any position or investment advice of Futu. Futu makes no representation or warranty.Read more
13K Views
Report
Comments
Write a Comment...
1