· reAPI Team
Who Makes Qwen? Alibaba Cloud's Qwen Models Explained
Qwen (千问) is made by Alibaba Cloud. Where the models come from, which are open weights, the 2026 release timeline, API prices, and Qwen vs DeepSeek.
Who makes Qwen? Alibaba Cloud. The official Qwen GitHub organization, QwenLM, describes itself as "Alibaba Cloud's general-purpose AI models" and links to qwen.ai, and it has been publishing models since August 2023. Its repository descriptions name the team directly: Qwen3-TTS, for example, is "developed by the Qwen team at Alibaba Cloud".[1] In Chinese, the same family is branded 千问 (Qianwen) on Alibaba Cloud's own Chinese-language site, which is why searches for "千问是哪个公司的" lead back to Alibaba.[2]
Qwen is also two things at once: a set of open-weight models anyone can download, and a set of hosted models sold through Alibaba Cloud Model Studio. Knowing which is which answers most practical questions about licensing, cost and where to run it.
Who makes Qwen, and where it is published
Three official channels carry Qwen releases:[1][3]
- GitHub (QwenLM): code, model information and issue tracking. Most repositories are Apache 2.0, including the current Qwen3.8 series repository.
- Hugging Face and ModelScope: the model weights. The Qwen README recommends ModelScope for users who cannot reach Hugging Face.
- Qwen Studio at chat.qwen.ai, the consumer chat app, linked from the same README.
The hosted API runs on Alibaba Cloud Model Studio, which lists model IDs such as qwen3.8-max, qwen3.8-flash and earlier qwen3.7, qwen3.6 and qwen3.5 versions.[4]
Qwen's 2026 release timeline
The Qwen3.8 repository lists the recent open releases:[3]
| Date | Release |
|---|---|
| February 16, 2026 | Qwen3.5, starting with a 397B-A17B mixture-of-experts model |
| February 24, 2026 | Qwen3.5-122B-A10B, 35B-A3B and 27B |
| March 2, 2026 | Qwen3.5 small models: 9B, 4B, 2B and 0.8B |
| April 16 and 22, 2026 | Qwen3.6-35B-A3B and Qwen3.6-27B |
| August 12, 2026 | Qwen3.8-2.4T-A95B |
| August 14, 2026 | Qwen3.8-27B |
Alibaba says Qwen3.8 "for the first time" brings a Qwen-Max-class model to open release, built on the Qwen3.5 architecture.[3] Alibaba describes models named this way, such as Qwen3.5-397B-A17B and Qwen3-Next-80B-A3B, as mixture-of-experts models, where the A figure is the share of parameters used for each token.
The README lists Qwen3.5's main changes as a unified vision-language base trained with early fusion, a hybrid architecture of Gated Delta Networks with sparse mixture-of-experts, and support for 201 languages and dialects. Qwen3.6 focused on agentic coding and keeping thinking context across turns, and Qwen3.8 adds reasoning_effort and preserve_thinking controls.[3]
More than chat models
The QwenLM organization covers far more than a chatbot. Its repositories include Qwen3-VL for vision-language work, Qwen3-Omni, Qwen3-TTS for speech generation and voice cloning, Qwen3-ASR for speech recognition, Qwen-Image for image generation and editing with complex text rendering, and Qwen3-Coder for code. Its most-starred repository is Qwen Code, "an open-source AI coding agent that lives in your terminal", released under Apache 2.0.[1]
Running Qwen open weights yourself
Several Qwen models are small enough for one machine. Ollama's library lists, for example, qwen3.6:27b at 17 GB and qwen3.6:35b at 23 GB in 4-bit form, both with a 256K context, and qwen3-coder:30b at 19 GB. The larger qwen3-coder-next is 52 GB.[7] The flagship Qwen3.8-2.4T-A95B is in a different class entirely and needs server hardware. Our guide to running a local LLM for coding sorts these by memory budget and explains why context length matters as much as model size.
What Qwen costs on the API
On Alibaba Cloud Model Studio's International pricing, qwen3.8-max costs $2 per million input tokens and $6 per million output tokens for requests up to 1 million tokens, in both thinking and non-thinking modes. The previous qwen3.7-max lists at $2.50 and $7.50.[4] The open-weight models cost nothing to download; running them means paying for your own GPUs.
Qwen vs DeepSeek
People often compare Qwen with DeepSeek, the other widely used Chinese model family. The practical differences, from each company's documentation:
| Qwen | DeepSeek | |
|---|---|---|
| Developer | Alibaba Cloud | DeepSeek |
| Open weights | Yes, Qwen3.5, 3.6 and 3.8 series, mostly Apache 2.0 | Check each model card |
| Hosted API | Alibaba Cloud Model Studio | DeepSeek's own API at api.deepseek.com |
| Flagship API price | qwen3.8-max: $2 / $6 per 1M tokens | deepseek-v4-pro: $0.66 / $1.98 off-peak, double at peak |
| Model range | Text, vision, omni, audio, image, coding | Text, with image input on Flash |
Sources: QwenLM GitHub and Alibaba Cloud pricing; DeepSeek models and pricing.[1][4][5] DeepSeek's prices change by time of day: its peak hours are 01:00 to 04:00 and 06:00 to 10:00 UTC on weekdays.[5] On list price alone DeepSeek is cheaper; Qwen covers more modalities and publishes more model sizes for self-hosting. Quality depends on the task, so test both on your own prompts. For DeepSeek's billing in detail, see our DeepSeek V4 context and pricing guide.
FAQ
Is Qwen open source?
Many Qwen models are open weights. The Qwen3.8, Qwen3.6 and Qwen3.5 series are published on Hugging Face and ModelScope, and the QwenLM repositories are mostly Apache 2.0. Check each model card for its exact license.[1][3]
Is Qwen a Chinese model?
Yes. Qwen is developed by the Qwen team at Alibaba Cloud, and its Chinese brand on Alibaba Cloud's site is 千问. Its weights are published on both Hugging Face and the Chinese platform ModelScope.[1][2][3]
Is Qwen the same as Tongyi?
Alibaba Cloud's Chinese site now uses the 千问 (Qianwen) brand for these models and apps, and the English name is Qwen.[2]
Where can I use Qwen?
In Qwen Studio at chat.qwen.ai, through Alibaba Cloud Model Studio's API, or by downloading the open weights from Hugging Face or ModelScope.[3][4]
Using Qwen models on reAPI
reAPI serves Qwen's image models rather than its language models. Qwen Image 3 costs $0.024 per image on the Standard tier and from $0.035 per image on Pro at 1K, checked on November 4, 2026.[6] Our guide to whether Qwen Image 3 is open source covers its weights and license. For Qwen's text models, Alibaba Cloud Model Studio or the open weights are the direct routes; reAPI's language models include DeepSeek V4 and Kimi K3.
So when someone asks who makes Qwen, the short answer is Alibaba Cloud's Qwen team, and the longer one is that it ships two ways: open weights you can host and audit, and hosted Model Studio models you pay for per token. Pick the route by where your data has to live and how much infrastructure you want to run.
References
- GitHub. QwenLM organization profile and repositories. Retrieved November 4, 2026 from github.com/QwenLM
- Alibaba Cloud (阿里云). 千问 product site. Retrieved November 4, 2026 from tongyi.aliyun.com
- Qwen. Qwen3.8 repository README (Qwen3.5, Qwen3.6 and Qwen3.8 series). Retrieved November 4, 2026 from github.com/QwenLM/Qwen3.8
- Alibaba Cloud. Model Studio: model pricing. Retrieved November 4, 2026 from alibabacloud.com/help/en/model-studio/model-pricing
- DeepSeek. Models & Pricing. Retrieved November 4, 2026 from api-docs.deepseek.com/quick_start/pricing
- reAPI. Qwen Image 3 model page. Retrieved November 4, 2026 from reapi.ai/models/qwen-image-3
- Ollama. Model tag pages: qwen3.6, qwen3-coder and qwen3-coder-next. Retrieved November 4, 2026 from ollama.com/library