Alibaba Cloud introduced ModelScope at the Apsara Conference with more than 300 models across language, vision, and audio. The platform offered a common place to download and fine-tune existing models.
Alibaba Cloud & Qwen
Also known as: Alibaba Cloud · Tongyi Lab · Qwen Team · ModelScope · Alibaba Cloud Model Studio · Bailian
Plain English
Alibaba's AI research and cloud platform division—operating the open-source ModelScope hub, the enterprise Alibaba Cloud Model Studio (Bailian) API platform, and the Tongyi research organization behind the Qwen language, coding, and multimodal family and the Wan video generation models.
Technical Overview
Alibaba pairs downloadable open weights on ModelScope and Hugging Face with commercial API hosting on Alibaba Cloud Model Studio (Bailian). Its Tongyi Lab trains dense and hybrid linear-attention MoE checkpoints across text, coding, vision, and audio (Qwen, Qwen-VL, Qwen-Coder, Qwen-Omni, Qwen-Image) alongside spatiotemporal DiT video foundation models (Wan), with availability and license terms varying by checkpoint.
Timeline (6)
2022
Ecosystem2023
ReleaseThe Qwen team released Qwen-7B and the instruction-tuned Qwen-7B-Chat on ModelScope and Hugging Face. Publishing both checkpoints gave developers a starting point for either further training or chat applications.
2025
ReleaseAlibaba’s team released Wan 2.1 weights under Apache 2.0, with 1.3B and 14B text-to-video models and a 14B image-to-video model. The smaller checkpoint made local experiments possible on hardware with less memory.
- Milestone
At the Apsara Conference, Alibaba Cloud announced new Qwen3 offerings and previewed the Wan 2.5 generation. The hosted Qwen3-Max endpoint and planned Wan updates broadened the text and visual capabilities available through Model Studio.
- Paper
The Qwen team published the Qwen3-VL technical report covering dense and MoE vision-language checkpoints that interleave text, documents, images, and long video inside a single context window, laying the early-fusion multimodal foundation later unified into Qwen3.5 (February 2026).
2026
ReleaseAlibaba Cloud introduced the hosted Qwen3.8-Max API and published weights for the Qwen3.8-2.4T-A95B model. The sparse checkpoint activates about 95 billion of its 2.4 trillion parameters per token.
See also (13)
Open full graphAlibaba Cloud & Qwen
8 more connections
- Tencent HunyuanOutgoing · Alternative to
- WanOutgoing · EnablesIncoming · Mentions
- Context Window & RoPEOutgoing · Mentions
- Diffusion Transformer (DiT)Outgoing · Mentions
- Hugging FaceOutgoing · Mentions
- Linear AttentionOutgoing · Mentions
- Mixture of Experts (MoE)Outgoing · Mentions
- Multimodal & Vision-Language ModelsOutgoing · Mentions