Harvey

Legal AI · Enterprise Software

Launch

Harvey launches Tenet, its first in-house AI model for legal work

OpenAI and Anthropic are now courting the same law firms that pay to use their models through this platform.

• Harvey introduced Harvey Tenet on Aug. 18, 2026, its first in-house proprietary AI model built specifically for legal work, moving beyond reliance on third-party models like OpenAI and Anthropic.
• Harvey, a legal AI startup valued at $11 billion, was co-founded by former Google DeepMind researcher Gabe Pereyra and Winston Weinberg.
• Tenet was built by fine-tuning Kimi K3, a low-cost open-source model from Chinese startup Moonshot, using legal reasoning data created by attorneys hired through Mercor and Snorkel to draft mock disputes and case files.
• The launch comes as rival AI labs push directly into legal work: OpenAI hired Ironclad founder Jason Boehmig to lead a legal push and Anthropic offers plugins for document review and drafting, with Google and Meta seen as possible entrants.
• Harvey has not disclosed Tenet's base model, parameter count, training data or benchmark scores, leaving lawyers to judge its reliability largely without published evaluations.
• As the AI labs supplying Harvey's models increasingly compete for its own customers, building an in-house model gives Harvey a hedge against supplier dependency and shrinking margins.