Title: OpenAI-Backed Startup Ditches US Models for Chinese Kimi K3

**Article:** A US artificial intelligence startup backed by OpenAI has built its first in-house AI model using a Chinese open-weight system, signaling a growing shift by Western tech firms toward cheaper Chinese alternatives. San Francisco-based legal tech company Harvey announced Thursday that its new model, Harvey Tenet, was trained on top of Kimi K3, an open-weight base model developed by Chinese lab Moonshot AI. Open-weight models allow developers to access and modify the underlying code, unlike closed systems such as OpenAI’s GPT-4. Harvey’s high-profile backers include OpenAI, Sequoia Capital, and Andreessen Horowitz. The company said the move was driven by soaring development costs for building models from scratch. By using Kimi K3 as a foundation, Harvey aims to cut expenses while tailoring the model for legal work, such as contract analysis and case research. The decision highlights a broader trend: US firms are increasingly turning to Chinese open-weight models as a practical, cost-effective alternative to developing proprietary systems. Harvey did not disclose the total cost of training Harvey Tenet, but said the approach reduced computing needs significantly compared to building a model from zero. The company emphasized that Harvey Tenet remains fully owned and controlled by Harvey, with all client data processed in the US. Kimi K3 was released by Moonshot AI earlier this year and has gained traction among developers seeking high performance without the high price tag of US-made models. Harvey said the new model will be rolled out to select clients in the coming weeks, with broader availability expected later this year.