Chinese AI 'Qwen3.5-397B-A17B,' equivalent to GPT-5.2, is now available as an open model



Qwen (Tongyi Lab), Alibaba's AI research team, announced the Qwen 3.5 series of AI models on February 16, 2026, and released the open model ' Qwen 3.5-397B-A17B ' as the first product in the series. The Qwen 3.5-397B-A17B demonstrates performance on par with closed models such as GPT-5.2, Claude Opus 4.5, and Gemini 3 Pro.

Qwen3.5: Towards Native Multimodal Agents

https://qwen.ai/blog?id=qwen3.5

GitHub - QwenLM/Qwen3.5: Qwen3.5 is the large language model series developed by Qwen team, Alibaba Cloud.
https://github.com/QwenLM/Qwen3.5?spm=a2ty_o06.30285417.0.0.72bcc921FFdCmR&file=Qwen3.5

Qwen3.5-397B-A17B is an AI model trained using trillions of tokens of training data, with a total of 397 billion parameters and 17 billion active parameters. It integrates the functions of large-scale language models and visual language models, and is characterized by its excellent coding and agent performance, as well as image and video understanding capabilities.

Qwen3.5-397B-A17B's performance improved with each reinforcement learning iteration. The final version of Qwen3.5-397B-A17B outperformed DeepSeek-V3.2-Thinking in the non-thinking domain, and outperformed closed models such as GPT-5.2 High, Gemini 3 Pro, and Claude Opus 4.5 in the thinking domain.



Below is a chart showing the results of various benchmarks measuring coding performance, visual comprehension performance, etc. From left to right, the scores for 'Qwen3.5-397B-A17B,' 'Qwen3-Max-Thinking,' 'Qwen3-VL-235B-A22B,' 'GPT-5.2,' 'Claude Opus 4.5,' and 'Gemini 3 Pro' are shown, with Qwen3-VL-235B-A22B recording the highest score in some tests.



The Qwen3-VL-235B-A22B boasts more efficient processing than previous Qwen series models, achieving 8.6 times faster processing speeds for 32K tokens and 19 times faster processing speeds for 256K tokens than the Qwen3-Max.



Below is a video recording a website being coded using the Qwen3-VL-235B-A22B. It shows a high-quality website being created.




In addition, we can also develop racing games using 3DCG.




The model data for Qwen3-VL-235B-A22B is available at the following link.

Qwen/Qwen3.5-397B-A17B · Hugging Face
https://huggingface.co/Qwen/Qwen3.5-397B-A17B?spm=a2ty_o06.30285417.0.0.72bcc921FFdCmR&file=Qwen3.5-397B-A17B



In addition to Qwen3-VL-235B-A22B, Qwen Chat can also use Qwen3.5-Plus, which supports input of 1 million tokens. Furthermore, Qwen3.5-Plus can also be used via the Alibaba Cloud Model Studio API.



Related Posts:

in AI, Posted by log1o_hf