In an ambitious move, Alibaba has introduced its Qwen 3.5 series, setting a competitive stance against renowned models like GPT-5 mini and Claude Sonnet 4.5, but at a significantly lower cost. Unveiled on February 26, 2026, the Qwen 3.5 lineup consists of four distinct models: Qwen3.5-Flash, Qwen3.5-35B-A3B, Qwen3.5-122B-A10B, and Qwen3.5-27B. These models promise enhanced performance while requiring less computational power, a feat that could redefine efficiency in AI development.
Each of these models is capable of processing text, images, and video inputs to generate text outputs. Notably, the compact Qwen3.5-35B-A3B model surpasses its larger predecessor, Qwen3-235B-A22B, underscoring the importance of advanced architecture, superior data quality, and reinforcement learning over sheer model size. Meanwhile, the 122B and 27B models are designed to bridge the performance gap, especially in complex agent scenarios.
Alibaba claims that the Qwen 3.5 models either match or exceed the capabilities of leading Western models, such as OpenAI's GPT-5 mini, gpt-oss-120b, and Anthropic's Claude Sonnet 4.5, across a range of benchmarks.
These models are accessible on platforms like Hugging Face, ModelScope, and Qwen Chat, under the Apache License 2.0. This open-source license permits commercial use, modification, and redistribution, broadening the models' appeal and usability.
The Qwen3.5-Flash model, designed for hosted production, features a context length of one million tokens and includes built-in tools. Its API pricing is structured at $0.10 per million input tokens and $0.40 per million output tokens, making it an attractive option for developers and businesses alike.