Moonshot AI’s Kimi K2 Model Surpasses U.S. Rivals in Breakthrough Open-Source Milestone
- Category: AI Automations
- Date 15-10-2025
A Chinese tech firm has notched an open source win.
Last week, Beijing-based startup Moonshot AI released its Kimi K2 Thinking model, its best open source model yet. The company claims the trillion parameter model excelled in major benchmarks for reasoning, agentic search, coding, writing and general capabilities.
The model surpassed proprietary competitors such as OpenAI’s GPT-5, Anthropic’s Claude Sonnet 4.5, and xAI's Grok-4 across several metrics. The model reportedly cost $4.6 million to train, CNBC reported.
The model is Moonshot’s second release, following its first, which debuted in July. The startup is backed by Chinese tech giants Alibaba and Tencent, and held a valuation of $3.3 billion after its most recent funding round last year of more than $300 million.
Moonshot’s model marks another successful open-source model from a Chinese company, following DeepSeek's market debut earlier this year, which achieved parity with US competitors’ models at a significantly lower cost. And in July, Alibaba-backed Z.ai released a powerful family of open source models, called GLM-4.5, which were able to undercut DeepSeek’s costs by 87%.
But Chinese tech firms aren’t the only ones with their focus on open-source AI. Last month, Reflection AI, which aims to challenge DeepSeek’s open-source prominence, announced a $2 billion funding round led by Nvidia, boosting its valuation to $8 billion. The company’s CEO and co-founder, Misha Laskin, told the New York Times that “there’s a DeepSeek-shaped hole in the U.S.”
However, while China and the US remain in a heated race to build powerful AI, Moonshot’s release might be the latest signal that China is edging ahead in open-source, affordable AI.