Moonshot AI’s Kimi K2 Thinking Model, reportedly trained for $4.6 million 🤯

Chinese AI lab Moonshot made waves earlier this year when its Kimi K2 model with 1 trillion parameters became one of the largest open-weight models worldwide. Now, Moonshot has released the “Thinking” version of K2. According to the company’s own benchmark results, K2 Thinking outperforms OpenAI’s GPT-5 and Anthropic’s Claude Sonnet 4.5 Thinking in Agentic Reasoning and Agentic Search tasks (https://simonwillison.net/2025/Nov/6/kimi-k2-thinking/).

Training cost? A source cited by CNBC (https://cnbc.com/2025/11/06/alibaba-backed-moonshot-releases-new-ai-model-kimi-k2-thinking.html) claims that the Kimi K2 Thinking model cost $ 4.6 million to train, though CNBC notes it could not independently verify the figures. The pace of competition in frontier AI continues to accelerate, not just in model size, but in reasoning capabilities and cost efficiency.