China’s Moonshot releases a new open source model Kimi K2.5 and a coding agent

China’s Moonshot AI, backed by Alibaba and HongShan, has released a new open source model called Kimi K2.5. This model understands text, images, and video. The company states it was trained on 15 trillion mixed visual and text tokens, making it natively multimodal. Moonshot AI also notes the model excels at coding tasks and handling agent swarms, where multiple AI agents work together in orchestration.

In released benchmarks, Kimi K2.5 matches the performance of proprietary models and even surpasses them in certain areas. For coding, it outperforms Gemini 3 Pro on the SWE-Bench Verified benchmark and scores higher than both GPT 5.2 and Gemini 3 Pro on the SWE-Bench Multilingual benchmark. In video understanding, it beats GPT 5.2 and Claude Opus 4.5 on the VideoMMMU benchmark, which measures a model’s reasoning over videos.

On the coding front, the model can understand text and also accept images or videos as input. Users can show it a media file and ask it to create a similar interface. To let people use these capabilities, the company has launched an open source coding tool called Kimi Code. This tool rivals offerings like Claude Code or Google’s Gemini CLI. Developers can use Kimi Code through their terminals or integrate it with development software such as VSCode, Cursor, and Zed. The startup confirms developers can use images and videos as input with Kimi Code.

Coding tools have become popular and significant revenue drivers for AI companies. Anthropic announced in November that Claude Code had reached one billion dollars in annualized recurring revenue, with reports indicating it added another one hundred million dollars by the end of 2025. Moonshot’s Chinese competitor, DeepSeek, is reportedly set to release a new model with strong coding abilities next month.

Moonshot AI was founded by former Google and Meta AI researcher Zhilin Yang. The company raised one billion dollars in a Series B round at a two and a half billion dollar valuation. According to reports, the startup secured an additional five hundred million dollars in funding last month at a four point three billion dollar valuation and is already seeking to raise a new round at a five billion dollar valuation.