Moonshot AI, a Chinese artificial intelligence company, announced that its Kimi chatbot can now handle up to 2 million Chinese characters in a single prompt, surpassing the capabilities of existing AI tools in China.
The Beijing-based startup, also known as Yuezhi Anmian, disclosed that this feature is currently in beta testing, available to invited users before a wider rollout. Launched in October, the Kimi chatbot is built on Moonshot AI’s self-developed Kimi large language model (LLM).
Previously, the Kimi chatbot could manage up to 200,000 Chinese characters in a context window, indicating the amount of text an AI model can process during interactions with users.
Xu Xinran, Moonshot AI’s engineering vice-president, stated, “We believe that the exponential expansion of the size of an LLM’s context window will help unleash users’ imagination for various AI applications.” These applications may involve understanding and analyzing complete code repositories or developing intelligent agents capable of autonomously performing multi-step tasks.
Large language models (LLMs) like Kimi are foundational to generative AI services such as ChatGPT, enabling the creation of diverse content types like audio, code, images, text, simulations, and videos.
Moonshot AI’s founder, Yang Zhilin, emphasized the importance of processing long prompts for the future advancement of AI models during an industry event last year.
With the updated feature of Kimi, Moonshot AI has surpassed the prompt capacity of Baichuan, a company established by Sogou founder Wang Xiaochuan. Last October, Baichuan2-192k LLM was acclaimed for handling approximately 350,000 Chinese characters in a context window.
This enhancement to Kimi’s prompt capacity further solidifies Moonshot AI’s position in the AI industry, following its record-breaking fundraising as the largest single financing raised by a Chinese AI startup since the release of OpenAI’s ChatGPT in November 2022.