Archive · 9 Nov 2025
Chinese AI Kimi K2 Thinking Outperforms ChatGPT and Claude on Key Benchmarks
Imagine a big competition where different computer programs, called AI chatbots, are tested to see how smart they are.
One new AI from China, named Kimi K2 Thinking, did really well on some of these tests, even beating very famous ones like ChatGPT and Claude.
For tests that check how well they understand questions and find information on the internet, Kimi K2 Thinking scored the highest.
However, when the tests were about writing computer code, Kimi K2 Thinking wasn't quite as good as ChatGPT.
This shows that AI from different countries are getting better and better, and it's becoming a close race.
Moonshot AI has released a new AI model called Kimi K2 Thinking.
Kimi K2 Thinking outperformed ChatGPT and Claude on the Humanity's Last Exam benchmark.
The Kimi K2 Thinking model also excelled in the BrowseComp benchmark for web browsing and search tasks.
In coding-related tasks measured by LiveCodeBench V6, Kimi K2 Thinking scored lower than ChatGPT and Claude.
The performance of Kimi K2 Thinking varied across benchmarks, with Gemini 2.5 Pro leading in LMArena's ranking.
- Who
- Moonshot AI's Kimi K2 Thinking
- What
- Outperformed ChatGPT and Claude on Humanity's Last Exam and BrowseComp benchmarks, though lagged in coding tasks.
- When
- This week (release of Kimi K2 Thinking)
- Where
- AI benchmarks (Humanity's Last Exam, BrowseComp, LiveCodeBench V6)
- Why
- Demonstrates advancements in Chinese AI models competing with Western counterparts.
Kimi K2 Thinking Performance
ChatGPT & Claude Performance
Humanity's Last Exam (HLE) Benchmark
Kimi K2 Thinking Performance
Kimi K2 Thinking scored 44.9%.
ChatGPT & Claude Performance
GPT-5 scored 41.7%, Claude Sonnet 4.5 Thinking scored 32%.
BrowseComp Benchmark (Web Browsing & Search)
Kimi K2 Thinking Performance
Kimi K2 Thinking garnered a score of 60.2%.
ChatGPT & Claude Performance
GPT-5 scored 54.9%, Claude Sonnet 4.5 Thinking scored 24.1%.
Coding Tasks (LiveCodeBench V6)
Kimi K2 Thinking Performance
Kimi K2 Thinking received a score of 83.1%.
ChatGPT & Claude Performance
GPT-5 scored 87%, Claude Sonnet 4.5 Thinking scored 64%.
- New Model Released
- Kimi K2 Thinking by Moonshot AI
- Humanity's Last Exam (HLE) Score
- Kimi K2 Thinking: 44.9%
- BrowseComp Score
- Kimi K2 Thinking: 60.2%
- LiveCodeBench V6 Score (Coding)
- Kimi K2 Thinking: 83.1%
- Competitors Mentioned
- ChatGPT, Claude, Gemini