Qwen3.6-35B-A3B released!

Alibaba's Q-AI research team has launched Qwen3.6-35B-A3B, a significant upgrade to their open-weight model series specifically engineered for efficient API deployment. The 35-billion parameter model demonstrates performance on par with OpenAI's GPT-4 on standard benchmarks like MMLU and HumanEval, while delivering a 3x inference speed improvement over the previous Qwen2.5-32B model. This leap in efficiency is achieved through advanced architectural optimizations and a new inference engine, making it one of the fastest models in its class.

Available immediately through Alibaba Cloud's DashScope API platform, Qwen3.6-35B-A3B offers developers a compelling alternative to more expensive closed models. The API pricing is approximately 70% lower than comparable offerings from major US providers, dramatically reducing the cost of building and scaling AI-powered features. With its 128,000-token context window and strong multilingual capabilities, the model is positioned for complex tasks like long-form document analysis, code generation, and multi-turn conversational agents.

The release underscores the intensifying competition in the global AI infrastructure market, where performance-per-dollar is becoming a critical battleground. By offering near state-of-the-art reasoning at a fraction of the cost, Alibaba is directly challenging the dominance of Western API providers and accelerating the commoditization of high-end AI capabilities. This move is likely to pressure other providers to lower prices and improve efficiency, benefiting developers and enterprises worldwide.