China is home to the world’s top artificial intelligence (AI) models that are open-sourced, according to an American benchmarking platform created by researchers from the University of California, Berkeley.
Kimi K2, MiniMax M1, Qwen 3 and a variant of DeepSeek R1 were ranked as the world’s top open-sourced AI models, beating out offerings like Google’s Gemma 3-72B and Meta’s Llama 4-Maverick, LMArena said in a report on Friday.
The winners were determined by “comparing them side by side and casting votes for the better response”, LMArena said on its website. The platform, previously known as Chatbot Arena, has been used by major AI companies like OpenAI and Google to assess their AI models.
LMArena highlighted the progress of Kimi K2, the top model on the list, which was launched by Chinese AI start-up Moonshot AI on July 11. Moonshot said it open-sourced two versions of Kimi K2.
On its X account on Friday, LMArena said Kimi K2 was “one of the most impressive” open-source large-language models (LLMs) to date, adding that it was gaining popularity because its user responses were “humorous without sounding too robotic”.
DeepSeek R1-0528, a fine-tuned version of the model launched by Hangzhou-based start-up DeepSeek at the beginning of the year, came in second as it was “strong in multi-turn dialogue and reasoning tasks”, the platform said.