Alibaba’s Qwen AI team has unveiled the Qwen3.5 Medium Model series, introducing four new large language models available for commercial use under an open-source license. These models, such as Qwen3.5-35B-A3B and Qwen3.5-122B-A10B, are now accessible to developers on platforms like Hugging Face and ModelScope. The key feature of these models is their impressive performance, outperforming well-known models like OpenAI’s GPT-5-mini and Anthropic’s Claude Sonnet 4.5 in benchmark tests.
The Qwen3.5 models stand out due to their sophisticated hybrid architecture, integrating Gated Delta Networks and a sparse Mixture-of-Experts system. The Qwen3.5-35B-A3B model, for instance, showcases parameter efficiency by activating only 3 billion out of its 35 billion parameters per token.
Alibaba has also released the Qwen3.5-35B-A3B-Base model to support the research community, and the Qwen3.5 lineup introduces a ‘Thinking Mode’ for internal reasoning chains. These models are optimized for various hardware environments, enabling organizations to leverage sophisticated AI capabilities without exorbitant costs.
By utilizing the Qwen3.5 Medium Models within private infrastructures, enterprises can enhance data handling and security while building reliable, autonomous agents with native tool-calling capabilities. This shift towards architectural efficiency over sheer scale ensures cost-conscious and agile AI integration, empowering organizations to meet evolving operational demands effectively.
Source: VentureBeat