Indonesian Political, Business & Finance News

Alibaba Releases New AI Model Qwen, Superior in Coding and More Efficient

| | Source: KOMPAS Translated from Indonesian | Technology
Alibaba Releases New AI Model Qwen, Superior in Coding and More Efficient
Image: KOMPAS

Alibaba, a Chinese technology company, has released the artificial intelligence (AI) model Qwen 3.6-35B-A3B. This model was launched two weeks after the release of Qwen 3.6 Plus in early April 2026. Qwen 3.6-35B-A3B is designed as an open-source AI model using a Mixture-of-Experts (MoE) architecture approach. As a result, this AI model only activates a portion of its parameters when in use. According to Alibaba, Qwen 3.6-35B-A3B is equipped with a total of 35 billion parameters. However, this AI model only activates 3 billion parameters in each computation process to make it more efficient. Qwen 3.6-35B-A3B is the successor to Qwen3.5-35B-A3B. Compared to its predecessor, Alibaba’s latest AI model is claimed to be superior in agentic coding. Its performance can even rival models with larger parameters, such as Qwen3.5-27B with 27 billion parameters and Gemma4-31B with 31 billion parameters. To prove its claims, Alibaba has shared benchmark scores for Qwen 3.6-35B-A3B. In benchmarks measuring the AI model’s capabilities in agentic coding, such as Terminal-Bench 2.0, Claw-Eval, and QwenWebBench, Qwen 3.6-35B-A3B outperforms Qwen3.5-27B and Gemma4-31B. In other tests like SWE-bench Verified and SWE-bench Pro, Alibaba’s new AI model rivals the scores of other large AI models. Besides coding, Qwen 3.6-35B-A3B also demonstrates strong performance in vision language, which is the ability to understand and process combinations of text and images simultaneously. In various vision-language benchmarks, its performance is claimed to match larger models like Claude Sonnet 4.5. In some tasks, Qwen 3.6-35B-A3B even surpasses the performance of AI models from Anthropic. The superiority of Qwen 3.6-35B-A3B is evident in aspects of spatial intelligence, or the ability to understand positions and relationships between objects in images, with a score of 92 on RefCOCO and 50.8 on ODInW13. Qwen 3.6-35B-A3B can be integrated with various popular coding assistants such as OpenClaw, Claude Code, and Qwen Code, to facilitate software development workflows with more efficient and contextual coding experiences. This AI model will also be available on Alibaba Cloud Model Studio in the near future.

View JSON | Print