Moonshot AI Launches Kimi K2, Says It's Better than GPT-5
Moonshot AI, a Chinese startup, has now released Kimi K2 Thinking, a new open language model that has gone public for testing. According to internal benchmarks, the new model is on par with or sometimes outperforms those such as GPT-5 and Claude 4.5.
Core Capabilities and Performance
Kimi K2 Thinking performs complex and multi-step operations automatically and has key abilities such as step-by-step reasoning and the use of external tools.
On benchmarks developed for high analytical skills, Kimi K2 passed record results in the "Humanity's Last Exam," a form of thousands of expert-level questions. Its respective strong point is in programming and retrieving information from the internet. In a free-form benchmark termed 'BrowseComp,' it posted above 60% while average-humans score under 30%. This indicates strong capacity in searching data and verifying facts as well as writing codes sequentially.
Future Development: Intelligent Agents
Kimi K2 Thinking from Moonshot AI is said to be the core technology for the company's future "intelligent agents." Over the long term, the company wishes to create these systems that do not simply answer questions but instead plan and somewhat execute complex tasks, from software development to scientific research.
How to Access Kimi K2 Thinking
The Kimi K2 Thinking model is currently live for public usage on the official Moonshot AI page.
