Harvey Launches Proprietary AI Model 'Tenet' to Compete With Its Own Suppliers

Harvey, the legal AI startup valued at $11 billion, has introduced its first in-house model called Harvey Tenet — a significant strategic pivot for a company that built its entire business running on top of OpenAI and Anthropic's models. The launch is part of a broader product update called Harvey II, which also includes a 'Memory' feature that lets agents retain user preferences across tasks.

The timing is no accident. Anthropic has been aggressively courting law firms with document review and drafting tools, while OpenAI poached Ironclad founder Jason Boehmig to spearhead its legal push. With Google and Meta also eyeing the space, Harvey faces an existential tension: its key technology suppliers are becoming direct competitors for the same customers Harvey has spent years cultivating.

Building a proprietary model addresses two problems at once. First, economics: every time a lawyer uses Harvey, the company pays inference fees to whichever third-party model handles the request — costs that compound rapidly at scale. Routing more queries through Tenet could meaningfully improve margins without raising prices. Second, quality: co-founder and former Google DeepMind researcher Gabe Pereyra argues that a model trained specifically on legal reasoning will outperform general-purpose alternatives on the work lawyers actually do. To build the training data, Harvey hired attorneys — through staffing platforms including Mercor and Snorkel — to construct mock disputes and grade model outputs.

Tenet itself is built on Kimi K3, a low-cost open-source model from Chinese startup Moonshot AI that has generated significant buzz since its July release for its performance-to-cost ratio. Harvey's benchmark results — due to be published soon — warrant healthy skepticism, since model makers routinely train on the same benchmarks they use to advertise performance. Tenet is not yet live in Harvey's product, and the company declined to name law firms currently testing it.

The longer-term ambition is more transformative than just cost savings. Pereyra envisions Tenet becoming a foundation model that individual law firms train on their own proprietary data — decades of negotiation experience, clause preferences, and deal-making instincts currently locked in attorneys' heads or buried in old documents. That would shift Harvey's business model from software vendor toward something closer to a Big Four consulting firm, configuring bespoke AI infrastructure around each client's unique institutional knowledge. For a company once dismissed as merely wrapping existing AI, that repositioning could prove to be its most durable competitive advantage.