Alibaba’s New Multimodal AI Model is Not Open-Source
multimodal open-source qwen
| Source: The Information | Original article
Alibaba Group has launched Qwen 3.5‑Omni, its latest large‑language model that can ingest text, audio, images and video, but this time the company is keeping the model proprietary. The shift marks a sharp departure from the open‑source stance Alibaba adopted with earlier releases such as Qwen‑3 and the September‑2023 Qwen‑3‑Omni, which made their weights publicly available.
The new model is offered only through Alibaba’s cloud AI services, where developers can access it via an API and pay per‑usage fees. Internally, the firm says the closed‑source approach lets it “ensure reliability, stability and rapid iteration” while protecting intellectual property that underpins its commercial offerings. The move follows the company’s aggressive monetisation push, exemplified by the agentic Qwen 3.6‑Plus announced just days earlier, which also arrived as a proprietary service.
Why it matters is twofold. First, Alibaba has been one of the few Chinese AI labs that contributed openly to the global research ecosystem; withdrawing Qwen 3.5‑Omni narrows the pool of high‑quality multimodal models available for academic benchmarking and community‑driven safety work. Second, the decision signals a broader industry trend where leading cloud providers are packaging advanced foundation models as revenue‑generating products rather than shared research artifacts, a pattern echoed by Microsoft’s recent in‑house model suite and Google DeepMind’s Gemma 4 rollout.
What to watch next includes Alibaba’s pricing strategy and any tiered access plans that could differentiate enterprise from startup users. Equally important will be the company’s response to community criticism—whether it will release limited‑scope checkpoints for research or double down on a fully closed ecosystem. The evolution of Qwen 3.5‑Omni will also serve as a bellwether for how Chinese AI firms balance open collaboration with the commercial imperatives of a rapidly maturing generative‑AI market.
Sources
Back to AIPULSEN