AI tools·Chat models·China
Qwen
Alibaba’s prolific open-weight family: chat, code, multimodal, and DashScope / Model Studio.
Compiled through 2026-09
Full open-weight range from 0.5B to MoE
- Vendor
- Alibaba / Alibaba Cloud
- Started
- 2023
- China availability
- Works in ChinaDirectly reachable from mainland China
- Pricing tier
- Free tier
- Current flagship
- Qwen2.5-Max / Qwen2.5-Coder
- Website
- tongyi.aliyun.com
Evaluation, Pros & Cons
Editorial score based on public benchmarks and community feedback — not a third-party benchmark result
Key Advantages
- Qwen 2.5-Max tops domestic Arena; rock-solid instruction following and Chinese nuance
- Full-spectrum open weights (0.5B to 72B & MoE) with top Coder variant for private hosting
Limitations & Caveats
- Web consumer UI is utilitarian; advanced tooling relies heavily on DashScope
Overview
Qwen is Alibaba’s model brand: open weights (Qwen → Qwen3) plus closed Max. April 2025’s Qwen3 moved to hybrid thinking; 2026 added 3.5 and 3.8, with Qwen3.8 Max / Flash-Next around Sep 2. Coder serves programming; Tongyi, DingTalk, and Quark are consumer surfaces. Apache-2.0 weights are a common private-deploy choice in China.
Direction
Alibaba Cloud is embedding Qwen in commerce, cloud dev, office, and agent platforms. Open weights and Coder will keep shipping; Max chases frontier reasoning. Multimodal and long context are explicit. Qwen4-class architecture is discussed in the community—wait for official releases.
Version history
Sep 2026
LatestQwen3.8 Max / Flash-Next
Flagship vs high-QPS split in early Sep.
View details
New capabilities
- Qwen3.8 Max flagship
- Flash-Next for high QPS
What got stronger
- Closed flagship and open family stay in parallel so you can split by job
Version notes
3.8 in early September split flagship vs high-QPS again. Max chases frontier reasoning; Flash-Next takes retrieval and support traffic. Treat Qwen4 as rumor until official.
May 2026
Qwen3.5
Plus / Flash value line.
View details
New capabilities
- 3.5 Plus / Flash value line
What got stronger
- High-QPS jobs could stay on Alibaba Cloud instead of fleeing to DeepSeek on price
Version notes
3.5 was mostly price and latency at scale. The intelligence bump was smaller than 3→3.8, but it mattered if you were already on Model Studio.
Apr 2025
Qwen3
Hybrid thinking; Coder-480B and more.
View details
New capabilities
- Hybrid thinking architecture
- Coder SKUs aimed at programming
What got stronger
- Open line started seriously chasing reasoning and code
Version notes
Qwen3 was an architecture break: thinking modes plus Coder. For developers that meant Tongyi was not only “decent Chinese prose” but IDE- and agent-ready.
Jun 2024
Qwen2
Open series upgrade, many sizes.
View details
New capabilities
- Qwen2 open weights in many sizes
- From small on-device to 72B-class
What got stronger
- Many more private-deploy options overnight
Version notes
Qwen2 made Tongyi one of the most prolific open families in China. Enterprises could pick a size for their GPUs instead of only buying Max.
Apr 2023
Tongyi Qwen launch
Alibaba Cloud chat + API brand.
View details
New capabilities
- Tongyi brand plus Alibaba Cloud API
- Chinese office and commerce from day one
What got stronger
- A billable, in-China cloud door to a large model
Version notes
Tongyi Qwen launched as Alibaba Cloud’s chat + API brand. Early quality was ordinary; distribution (cloud, DingTalk, later Quark) was the long game.