San Francisco-based legal AI firm Harvey, backed by OpenAI and top venture capital, has built its new in-house model using Chinese AI lab Moonshot AI’s open-weight Kimi K3 system. This marks a significant strategic shift amid rising AI development costs and highlights growing Western interest in Chinese AI technologies.
- Harvey Tenet built on Chinese Moonshot AI Kimi K3 open-weight model
- Achieved state-of-the-art results on complex legal tasks vs US models
- Open-weight approach lowers AI inference costs and enables custom post-training
What happened
Harvey, a US legal tech startup supported by OpenAI, Sequoia Capital, and Andreessen Horowitz, announced it has developed its first proprietary model on top of the Chinese AI lab Moonshot AI’s Kimi K3 open-weight foundation. Their new system, Harvey Tenet, was created through post-training on the Kimi K3 base over several months, aiming to enhance legal AI capabilities.
This move represents a pivot from Harvey’s previous approach that relied on adapting proprietary models from US AI leaders like Anthropic, OpenAI, and Google. By leveraging an open-weight Chinese model, Harvey reports achieving improved accuracy on complex, long-term legal tasks, as well as enhanced cost efficiency, using substantial GPU resources for training.
Why it matters
Harvey’s adoption of a Chinese open-weight model signals a notable shift in Western AI startups embracing non-US foundational systems amid the escalating costs and difficulties of developing advanced models independently. Open-weight models offer greater flexibility by allowing developers to post-train with specialized datasets, yielding more precise and application-specific results.
According to AI policy experts, this trend may expose the US’s lag in frontier open-weight model development, while also raising important considerations about the risks and strategic implications of integrating foreign AI technology into critical professional sectors such as law. Meanwhile, major companies like AT&T are cautiously evaluating these Chinese systems for enterprise use.
What to watch next
The broader legal tech industry will observe how Harvey’s open-weight AI approach influences adoption of non-US foundational models and whether it drives competitors to explore similar integrations. The cost and performance benefits claimed could accelerate more tailored, domain-specific AI applications across law and other professional services.
Meanwhile, regulatory, geopolitical, and security assessments of Chinese open-weight AI models will continue to play a key role in their acceptance by Western enterprises and governments. Market attention will focus on further advancements from Moonshot AI and competing open-weight systems, alongside evolving post-training techniques that maximize efficiency and accuracy.