Today at 1:40 PM
Moonshot AI has introduced Kimi K2 Thinking, its newest open-source AI model designed to function as a thinking agent. This release places Kimi K2 Thinking among the leading options for researchers and developers seeking state-of-the-art performance in reasoning, agentic search, coding, and writing tasks. The model achieves top results on benchmarks such as Humanity’s Last Exam and BrowseComp, demonstrating improved general capabilities.
Alongside these advances, Kimi K2 Thi…
Today at 1:40 PM
Moonshot AI has introduced Kimi K2 Thinking, its newest open-source AI model designed to function as a thinking agent. This release places Kimi K2 Thinking among the leading options for researchers and developers seeking state-of-the-art performance in reasoning, agentic search, coding, and writing tasks. The model achieves top results on benchmarks such as Humanity’s Last Exam and BrowseComp, demonstrating improved general capabilities.
Alongside these advances, Kimi K2 Thinking supports autonomous execution of 200 to 300 sequential tool calls without human involvement, reasoning coherently across hundreds of steps to address complex problems. This agentic workflow highlights the model’s ability to sustain logic and planning over extended interactions.
Building on these features, Moonshot AI continues its work in test-time scaling by increasing both the number of thinking tokens and the steps for tool use, a technical direction that underpins K2 Thinking’s advances. For those wanting early access, K2 Thinking is now available for use via kimi.com’s chat mode and through the Kimi K2 Thinking API. The full agentic mode will be released soon.
Today by Paul