Alibaba today released its latest AI model, based on the Qwen3-Next architecture, to the public. According to the developers, it is 10 times more powerful and costs 10 times less than the previous version.
Alibaba today released its latest AI model, based on the Qwen3-Next architecture, to the public. According to the developers, it is 10 times more powerful and costs 10 times less than the previous version.
Compared to its predecessor, the Qwen3-32B model, which was released in April as part of the Qwen3 family, the new Qwen3-Next-80B-A3B model with 80 billion parameters cost about ten times less to train and worked 10 times faster on some tasks, writes the South China Morning Post.
The developers also claim that the new model matches the performance of the company’s flagship model, the Qwen3-235B-A22B.
The company also unveiled a version of the model for logical computing based on the Qwen3-Next architecture. According to third-party tests, the Qwen3-Next-80B-A3B-Thinking outperformed both the company’s own Qwen3-32B-Thinking model and Google’s Gemini-2.5-Flash-Thinking model in several areas.
The new models are optimized to work on common devices, making them accessible to a wider range of users.
The Qwen3-Next-80B-A3B model is now available on Alibaba Cloud for API use; as well as on Hugging Face and ModelScope.
