Alibaba представляет Qwen-3.5: новая эра открытых ИИ-моделей с мультимодальностью и высокой эффективностью Translation: Alibaba unveils Qwen-3.5: a new era of open AI models with multimodality and high efficiency

The cloud division of the Chinese tech giant Alibaba has unveiled a new open-source AI model named Qwen-3.5. This was reported by SCMP.

Two versions of the large language model (LLM) have been made available in the Qwen application, coinciding with the Lunar New Year celebrations.

The flagship model showcases performance on par with leading advanced solutions, featuring a context window of 1 million tokens—one of the largest figures in the industry.

Qwen-3.5-Open-Source boasts 397 billion parameters and has a context window of 256,000 tokens.

For the first time, Qwen has achieved native multimodality, allowing it to comprehend text, images, audio, and video within a single system.

The family utilizes a new architecture aimed at enhancing computational efficiency and reducing operating costs.

Alibaba has stated that Qwen 3.5 is 60% cheaper to use and performs eight times better in handling heavy loads compared to its predecessor. It can also autonomously execute tasks in both mobile and desktop applications.

“Designed for the era of agent-based AI, Qwen 3.5 empowers developers and companies to accelerate their processes and achieve more with the same computational resources, setting a new benchmark for efficiency per unit of inference cost,” the company stated.

Qwen has gained significant popularity due to the release of multiple versions catering to various developer and business needs. The LLM has established itself as the “standard for open models.” In December, the number of downloads on Hugging Face surpassed the total downloads of all other major open solutions.

As a reminder, in January, Qwen-3 became the first-ever AI model to be uploaded and functioning in orbit.