Alibaba Group Holding has unveiled the Qwen3 series, the latest version of its flagship AI model, keeping pace with the rapid development in China’s AI sector following the rise of DeepSeek.

Announced on April 29, Qwen3 reportedly matches or surpasses DeepSeek’s performance in key areas such as mathematics and coding.

Alibaba also ssid that the new series significantly reduces deployment costs compared to other leading AI models.

The Qwen3 series includes two mixture-of-experts (MoE) models, designed to rival the hybrid reasoning systems recently introduced by Anthropic and Google. These models aim to emulate human-like problem-solving, enhancing the model’s ability to reason across complex tasks.

DeepSeek and other AI developers have adopted the Mixture of Experts (MoE) approach, which breaks down tasks into smaller data segments—similar to assigning different parts of a job to specialists—boosting overall efficiency.

After DeepSeek introduced a powerful model reportedly built for just a few million US dollars, Chinese tech companies responded swiftly, releasing a wave of affordable AI services that are rapidly transforming the market.

In 2025, after committing fully to the AI race, Alibaba launched a new model in its Qwen 2.5 series capable of processing text, images, audio, and video — and efficient enough to run directly on phones and laptops.

The company also introduced an updated version of its Quark AI assistant app in March.

OpenAI, Google, and Anthropic have also launched a wave of new models in recent months.

OpenAI recently announced plans to release a more “open” model designed to better mimic human reasoning, signaling a strategic shift following the open-source moves by DeepSeek and Alibaba.

On Tuesday, Alibaba confirmed that all its Qwen3 models are now open source.