Harvey, one of the most prominent AI legal startups backed by the OpenAI Startup Fund, is now incorporating foundation models from Google and Anthropic into its platform. This move marks a significant expansion beyond its original reliance on OpenAI technologies, as announced in a recent blog post.
While Harvey continues to collaborate closely with OpenAI, it has now added models from competitors to better serve the complex demands of legal work. The company’s decision follows internal testing through its proprietary benchmark, BigLaw, which revealed that no single model excels at all legal tasks. Instead, different models show strengths in specific areas. For example, Google’s Gemini 2.5 Pro performs exceptionally in legal drafting but struggles with tasks involving intricate legal reasoning like evidentiary rules. Meanwhile, OpenAI’s o3 and Anthropic’s Claude 3.7 Sonnet demonstrated stronger performance in pre-trial work.
Founded in 2022 with backing from OpenAI’s Startup Fund—then overseen by Sam Altman—Harvey has grown rapidly, reaching a $3 billion valuation as of February 2025. The startup’s recent $300 million Series D round was led by Sequoia, with participation from Coatue, Kleiner Perkins, and previous investors like GV and OpenAI. Google’s venture arm, GV, had also led Harvey’s $100 million Series C in mid-2024.
Instead of developing its own foundation models, Harvey’s strategy now focuses on integrating high-performing models from leading vendors and fine-tuning them for legal applications. This approach enables Harvey to build AI agents tailored to specific legal tasks, enhancing both flexibility and efficiency.
In addition, Harvey announced plans to launch a public leaderboard that tracks legal-specific AI benchmark results, accompanied by in-depth expert analysis. These efforts aim to add transparency to the competitive AI landscape and highlight the nuances of model performance that go beyond simple scoring.
