China’s Moonshot AI, backed by Alibaba, HongShan (formerly Sequioa China) and others, today released Kim K2.5, a new open source model that understands text, images, and videos.
The company says the model is trained on 15 trillion mixed visual and text tokens, which is why it’s natively multimodal. He further added that the model is good at coding tasks and handling swarms of agents.
— Orchestration where multiple agents work together. In released benchmarks, this model matches the performance of its own peers and even outperforms them on certain tasks.
For example, in coding benchmarks, Family K2.5 outperforms Gemini 3 Pro in the SWE-Bench Verified benchmark, and scores higher than GPT 5.2 and Gemini 3 Pro in the SWE-Bench Multilingual benchmark. In video understanding, it outperforms GPT 5.2 and Claude Opus 4.5 in VideoMMMU (Video Massive Multi-discipline Multimodal Understanding), a benchmark that measures how models reason about videos.
On the coding side, Moonshot AI says that while the model can understand text well, users can also feed the model images and videos and ask it to create an interface similar to what appears in those media files.
To give people access to these coding capabilities, the company launched an open-source coding tool called Kimi Code, which is comparable to Anthropic’s Claude Code and Google’s Gemini CLI. Developers can use Kim Code through the terminal and integrate it with development software such as VSCode, Cursor, and Zed. The startup said developers can use images and videos as inputs for Kim Code.
Coding tools are rapidly becoming a revenue driver for AI labs. Anthropic announced in November that Claude Code had reached $1 billion in annual recurring revenue (ARR). Earlier this month, Wired reported that the tool increased that number by $100 million by the end of 2025. Moonshot’s Chinese competitor Deepseek plans to release a new model with powerful coding technology next month, The Information reports.
Moonshot was founded by former Google and meta-AI researcher Yang Zhilin. The company raised $1 billion in Series B funding at a valuation of $2.5 billion. The startup raised $500 million in funding last month at a valuation of $4.3 billion, according to Bloomberg. Additionally, the report noted that the company is already looking to raise new funding at a valuation of $5 billion.
