Alibaba's Qwen3.5-Omni Is Proprietary, Marking a Shift Away From Its Open-Source Strategy

Alibaba's newly released Qwen3.5-Omni multimodal model is proprietary rather than open-weight, marking a notable departure from the company's previous strategy of releasing Qwen models for public download and use, according to a report by Juro Osawa at The Information cited by Techmeme on Tuesday. The shift is significant because Alibaba's Qwen series had established itself as a leading open-source AI model family, widely downloaded on Hugging Face and used by developers, researchers, and companies building AI applications who wanted capable models without API dependency or usage costs. Moving to a proprietary model reverses that positioning for at least this generation of the multimodal product. The reasons for the shift were not specified in initial reports. Companies that open-source AI models typically do so to drive ecosystem adoption, accelerate research, and build goodwill with developers. Moving to proprietary access could reflect a decision to monetize the model's capabilities more directly through API charges, or concern that open-weight distribution of a highly capable omnimodal system carries safety or competitive risks. Qwen3.5-Omni supports text, audio, image, and video inputs and reportedly surpasses Google's Gemini 3.1 Pro on audio benchmarks. The model supports more than 10 hours of continuous audio input, positioning it for enterprise transcription and analysis use cases. The proprietary turn comes as multiple AI companies have re-evaluated open-source strategies. OpenAI has not released open weights for its frontier models, while Meta has maintained its Llama series as open-weight. Google released Gemma as an open-weight complement to its proprietary Gemini family. Ali Cloud's API pricing for Qwen3.5-Omni was not immediately disclosed.