China's Kimi K2 Thinking Model Surpasses Leading AI Competitors

Moonshot AI, a Chinese startup founded in March 2023 by Yang Zhilin, has launched its Kimi K2 Thinking model, which has reportedly outperformed OpenAI's GPT-5 and Anthropic's Claude Sonnet 4.5 across several key benchmarks. Released on November 6, 2025, Kimi K2 Thinking achieved a score of 44.9% on Humanity’s Last Exam, surpassing GPT-5's performance. The model also excelled in coding tasks, scoring 83.1% on LiveCodeBench V6, and 60.2% on the BrowseComp benchmark, which evaluates agentic search capabilities.

China’s Moonshot AI Claims Model Beats GPT-5China’s Open Kimi K2 Thinking Model Beats GPT-5, Sonnet 4.5 On Humanity’s Last Exam & Agentic Tasks, Tops BenchmarksChina’s Open Kimi K2 Thinking Model Beats GPT-5 And Sonnet 4.5? Exponential View #549: Volatile markets; infinite compute; Kimi K2’s frontier leap; orbital computing++

Kimi K2 Thinking utilizes a mixture-of-experts architecture with a trillion parameters, enabling it to perform complex tasks autonomously. It can execute between 200 and 300 sequential tool calls without human intervention, showcasing its advanced reasoning capabilities. The model features a 256K context window and is designed for test-time scaling, enhancing its ability to handle extensive operations. Currently, it is available in chat mode on kimi.com, with full agentic mode and API access expected soon.

China’s Open Kimi K2 Thinking Model Beats GPT-5, Sonnet 4.5 On Humanity’s Last Exam & Agentic Tasks, Tops Benchmarks

The emergence of Kimi K2 Thinking signifies a potential shift in the global AI landscape, challenging the dominance of closed models from leading companies. Historically, the most capable AI systems have been proprietary, limiting access to their architectures and weights. The success of Kimi K2 Thinking suggests that open models may now rival these proprietary systems, potentially accelerating innovation and reducing reliance on commercial APIs. This development could empower researchers and enterprises to customize and deploy AI solutions independently.

China’s Open Kimi K2 Thinking Model Beats GPT-5 And Sonnet 4.5

While the performance claims of Kimi K2 Thinking are significant, independent verification of these results is still pending. Analysts caution that the competitive landscape is evolving rapidly, and the implications of this shift may not be fully understood yet. Concerns have also been raised regarding the broader impact of China's advancements in AI technology, particularly in the context of U.S.-China technological competition.

China’s Open Kimi K2 Thinking Model Beats GPT-5, Sonnet 4.5 On Humanity’s Last Exam & Agentic Tasks, Tops Benchmarks

Experts in the field, including NVIDIA CEO Jensen Huang, have expressed concerns about China's rapid progress in AI, emphasizing the need for the U.S. to enhance its efforts to remain competitive. Former Google CEO Eric Schmidt and venture capitalist Marc Andreessen have echoed similar sentiments, framing the race for AI supremacy as a geostrategic imperative for the United States.

China’s Open Kimi K2 Thinking Model Beats GPT-5, Sonnet 4.5 On Humanity’s Last Exam & Agentic Tasks, Tops Benchmarks

As Kimi K2 Thinking gains traction, its open-source nature may lead to increased adoption and influence in the AI community. The model's release could set new technical standards and foster collaboration among researchers and developers. The ongoing developments in AI, particularly from Chinese firms like Moonshot AI, will likely continue to shape the competitive landscape in the coming years.

China’s Open Kimi K2 Thinking Model Beats GPT-5, Sonnet 4.5 On Humanity’s Last Exam & Agentic Tasks, Tops Benchmarks