Moonshot AI Unveils Kimi K2 Thinking Model Following $500M Funding Round

Moonshot AI, a leading member of the Sky9 Capital portfolio, has announced the release of Kimi K2 Thinking, its most advanced open-source reasoning model. This release coincides with a successful 500 million dollar funding round designed to fuel the company's future research and expansion. The model represents a breakthrough in making high-level reasoning logic accessible to the global developer community.

The Kimi K2 Thinking model is built as an active thinking agent that interleaves logical reasoning with tool usage. It is capable of executing up to 300 sequential tool calls autonomously to solve intricate, multi-step problems. This allows the system to continuously browse, search, and refine its hypotheses without requiring human guidance during the execution process.

Performance evaluations show that the model sets new records in reasoning and agentic search. It achieved a state-of-the-art score of 44.9 percent on the Humanity’s Last Exam benchmark and surpassed human baselines on BrowseComp with a 60.2 percent result. These metrics highlight the model's ability to manage information-rich environments and complex academic inquiries with high precision.

On the technical side, the model employs a Mixture-of-Experts structure featuring 1 trillion parameters. Moonshot AI utilized specialized quantization techniques to achieve a 2x generation speed improvement on standard hardware. This efficiency ensures that the model can handle long-horizon tasks like coding and creative writing with minimal delay.

Looking ahead, Moonshot AI intends to use its new capital to refine agentic workflows for professional and research applications. The company is moving toward a future where AI acts as a reliable partner in software engineering and deep scholarly analysis. Developers can currently access the model through the Kimi API or the official chat interface at kimi.com.