ZUG/HANGZHOU—Recently, Alibaba Group’s release of the Qwen 3.5 model family drew increased attention across AI-focused social media and developer channels, with posts highlighting its scale, multimodal features, and benchmark comparisons to leading proprietary systems.
Alibaba and the Qwen team said the flagship open-weights model, Qwen3.5-397B-A17B, uses a mixture-of-experts design with 397 billion total parameters and 17 billion active parameters per token. The company also described the series as “native multimodal,” targeting agentic use cases that involve interacting with applications through visual inputs.
The company’s materials and third-party reporting also pointed to expanded long-context support. Alibaba has promoted a hosted offering, Qwen3.5-Plus, with a context window of up to one million tokens, while the base open model is generally described as supporting a smaller native context window.
Alibaba said Qwen 3.5 is designed to execute multi-step tasks with lower inference costs and improved throughput versus its previous flagship, and it highlighted “visual agentic capabilities” that allow the model to take actions across mobile and desktop applications. Reuters reported Alibaba’s claim that Qwen 3.5 is cheaper to use and more capable at processing large workloads than its predecessor, and that the company’s benchmark disclosures compare the system with U.S. models.
Coverage by the South China Morning Post described Qwen-3.5 as Alibaba Cloud’s next-generation open model series and noted that Alibaba’s comparisons rely on benchmark results that are, in part, self-reported by the company.
GlobalData, citing its social media analytics, said influencer commentary focused on technical elements such as decoding speed claims, multilingual support, multimodal/agent capabilities, and the feasibility of running some variants on high-end personal hardware. The same analysis characterized the release as part of an intensifying competition among Chinese and U.S. model developers. (GlobalData’s specific measurements and selections of posts were not independently verified in this article.)
Several widely shared posts cited by GlobalData framed Qwen 3.5 as an attempt to maintain competitiveness amid other recent open-model releases, while others emphasized benchmark charts and cost/performance claims. A separate wave of discussion centered on deployment options—ranging from hosted APIs to community tooling—and how long-context and multimodal features could affect agent-style applications.
