Legal software company Harvey has unveiled Harvey Tenet, its first in-house AI model built specifically for legal tasks. The model was trained on mock disputes and case files using a customized version of Kimi K3.
Harvey, valued at $11 billion, built its business by layering products on top of third-party AI models from companies like OpenAI. Harvey Tenet marks the company's first attempt to develop proprietary AI technology in-house.
The model was trained using mock disputes and case files, giving it specialized legal knowledge beyond general-purpose AI systems. Harvey used a version of Kimi K3—likely referencing technology from Chinese AI firm Deepseek or a similar model—as its foundation.
The move signals Harvey's ambition to reduce reliance on external AI providers and build defensible competitive advantages. Legal-focused AI models require domain-specific training data and fine-tuning to handle the nuances of legal reasoning, contract analysis, and dispute resolution.
Harvey competes in a crowded market of AI-powered legal tools. Other players have similarly invested in proprietary models tailored to legal work, recognizing that off-the-shelf AI lacks the specialized training necessary for high-stakes legal applications.
Developing in-house models carries risks. Training requires substantial computational resources and curated legal datasets. Quality control matters—errors in legal AI can have costly consequences for clients.
The announcement comes as legal tech companies race to demonstrate AI capabilities that justify premium pricing. Harvey's existing $11 billion valuation depends partly on proving that its platform delivers unique value. A proprietary model could strengthen that narrative.
Harvey Tenet's performance in real-world legal workflows will determine its success. The company will need to show that its specialized training produces better results than general-purpose models for specific legal tasks like contract review, due diligence, or case analysis.
Artificial Analysis released the "Search Index," a benchmark comparing search API providers on quality, cost, and speed for AI agents. Luna, Parallel, Exa, and Firecrawl ranked highest among seven tested providers.
Snowflake CEO Sridhar Ramaswamy outlined a strategy to reduce enterprise AI expenses by moving beyond the assumption that bigger models always deliver better results. The approach prioritizes economic efficiency and risk mitigation in AI deployment.
Perplexity saw revenue in India surge 60% after ending its free AI offer through Airtel, despite a decline in new downloads. The shift suggests the company successfully converted free users into paying customers.