Alibaba Targets Coding and Office Tasks With Low-Cost AI Model

Alibaba has unveiled Qwen3.8-Flash, a new artificial intelligence model aimed at delivering cost efficiency, according to a post on social media platform X on August 26, 2026. This multimodal mixture of experts (MoE) model, set to be available soon via the QwenCloud API, offers an impressive price point of $0.16 per 1 million input tokens and $0.47 per 1 million output tokens.

The production version boasts 125 billion parameters and 51 billion N-gram embeddings, with 6 billion activated per token, making it 1/9th the cost of the previous Qwen3.7-Plus model while outperforming it across various tasks, particularly in coding and office applications. Alibaba claims this significant cost reduction is achieved without compromising performance, and their AI models have become the world's leading offerings, with over 3 billion downloads worldwide in the past six months.

The company has open-sourced more than 460 models, and its ecosystem has generated over 300,000 derivatives. Apple recently collaborated with Alibaba to develop an AI model tailored for the Chinese market, which will be integrated into Apple's AI suite, Apple Intelligence, set to launch in China soon. This local model will provide Apple with greater control over the AI experience offered in the country, potentially becoming the first foreign company to receive Chinese government approval for a proprietary model.