Alibaba's Qwen team has introduced Qwen3.8-Flash-Next, a new mixture-of-experts AI model that activates only 6 out of 125 billion parameters per token, significantly enhancing cost efficiency and performance, according to The Decoder.
The Decoder reported that this model operates at just one-ninth the training cost compared to traditional models while outperforming larger competitors such as DeepSeek-V4-Flash and Anthropic's Claude Opus 4.6 on coding and office task benchmarks.
This advancement highlights Alibaba's push in AI innovation, which could influence Japanese markets by accelerating local adoption of cost-effective AI technologies in sectors like finance and software development.