Chinese AI company Moonshot AI unveiled its new Kimi K2 model on July 16, 2023, which outperforms OpenAI's GPT-4 in several key performance metrics while being completely free to use. The company, which has previously developed acclaimed AI models, has achieved a significant breakthrough for Chinese AI technology's international competitiveness. Kimi K2 delivers particularly impressive performance in coding and mathematical tasks, scoring 80.3% and 77.9% in benchmarks, compared to GPT-4's 80.2% and 76.6% respectively.
Moonshot AI, backed by Alibaba Group, reached this milestone just 8 months after its founding in November 2022, and had secured a $1.5 billion valuation by July 14, 2024. The Kimi K2 model utilises 33.2 billion parameters and features Chinese, English, and coding capabilities, significantly enhancing its versatility. According to Yang Li, the company's co-founder and CEO, Kimi K2 pushes new boundaries in generative AI, with the model being particularly strong in handling Chinese language content, where it performs 25% better than GPT-4. Moonshot AI's open-access business model presents a strong challenge to paid AI services, especially in the Chinese market, where the company competes with the C2C AI Chat Platform contender alongside other strong local competitors such as Baidu and Zhipu AI.
Kimi K2's performance is remarkable as it achieves 12.3% better results in the GSM8K mathematics benchmark than Claude 3 Opus and performs 16.7% better in the MBPP coding test than Claude 3 Haiku. Moonshot AI's models offer free access both through API and web interface, representing a significant advantage over paid models like GPT-4, which requires a $20 monthly subscription. The company's strategy of taking an open-source approach aligns with the Chinese government's efforts to strengthen the country's position in the global AI race, particularly in open-source AI development, where American Meta and French Mistral AI currently dominate with their Llama and Mistral models.
Sources:
Michael Nuñez
Wency Chen
