hinese AI firms continue to prove themselves as open-source champions.

On Tuesday, Beijing-based Moonshot AI released Kimi 2.5, dubbing it the “most powerful open-source model to date.” The model is the latest iteration of the Alibaba-backed startup’s flagship model, claiming that it’s capable of processing text, images and video simultaneously.

Moonshot said in its announcement that Kimi 2.5 performs on par or surpasses closed-source rival models from Anthropic, OpenAI and Google in benchmarks related to video, image, coding and agents.

• Kimi 2.5 can also self-direct an “agent swarm” of up to 100 sub-agents, executing complex tasks in parallel across 1,500 tool calls, according to the company.
• Additionally, Moonshot AI claims its model is the “strongest open-source model to date for coding,” with particular strength in front-end development.

Moonshot AI’s success isn’t the only recent open-source win from a Chinese firm. On Tuesday, DeepSeek revealed an upgraded version of its optical character recognition model, a 3-billion parameter model that achieves state-of-the art performance for vision and document understanding.

These achievements are the latest sign that China is making significant open-source progress. Models like Alibaba’s Qwen continue to climb the ranks in popularity as developers search for cheaper and more efficient alternatives to proprietary, closed-source AI.