Moonshot AI Launches Kimi Chat, Supporting Input of 200,000 Chinese Characters


The main battlefield of large model services is shifting toward retrieval-augmented question answering, multi-document summarization, and long-range agents. The request prompt is composed of dozens to hundreds of semantically independent segments (retrieved documents, skills explanations, memory, historical rounds), forming ultra-long context with tens of thousands to hundreds of thousands of tokens. The pre-filling stage dominates the computing costs, becoming the most prominent cost source for service providers, and triggering more challenging problems.
The well-known AI startup "Moonshot AI" is seeking another round of financing, targeting $2 billion. If successful, its valuation could reach $30 billion, a sevenfold increase from last December. The company has developed the chatbot Kimi and is in discussions with potential investors, expecting to raise at least $1 billion.
Moonshot AI officially announced that the API for the Kimi K2 series model will cease maintenance on May 25th. Users are required to migrate to the latest model, kimi-k2.6, to obtain ongoing support and enhanced multimodal capabilities. The K2 series includes multiple versions, which were renowned for their trillions of parameters since their release in July of last year and have now reached the end of their lifecycle.
After completing a $2 billion funding round, Moonshot AI initiated an organizational restructuring, dismantling the VIE and red chip structure to remove regulatory obstacles for its Hong Kong IPO. The core of the restructuring involves converting offshore entities into domestic joint ventures, ensuring compliance with regulations for Chinese technology companies listed overseas, paving the way for the Hong Kong listing.
Moonshot AI has released and open-sourced its flagship model Kimi K2.6, achieving significant improvements in programming, long-range tasks, and multi-agent collaboration. The model is now available on the company's official website, app, API, and programming assistant, and has demonstrated excellent performance in multiple authoritative tests, including the challenging Humanity's Last Exam benchmark.