Harvey, a fast-growing legal AI startup backed by the OpenAI Startup Fund, has announced it will begin using foundation models from Anthropic and Google in addition to OpenAI’s. This marks a significant shift for the $3 billion company, which until now relied exclusively on OpenAI’s technology. Harvey emphasized that it is not replacing OpenAI but expanding its model options to improve performance and flexibility.
This decision is especially noteworthy given Harvey’s close relationship with OpenAI. The company was one of the first four startups funded by the OpenAI Startup Fund in 2022, when Sam Altman was still managing the fund. Despite this backing, Harvey has chosen to diversify its model stack, reflecting broader trends in the AI space where startups prioritize results over brand loyalty.
According to Harvey, its internal benchmarking system, BigLaw, revealed that multiple foundation models now outperform its original system on key legal tasks. These benchmarks showed that different models excel in different areas—Google’s Gemini 2.5 Pro is strong in legal drafting but weaker in pre-trial argumentation, while OpenAI’s o3 and Anthropic’s Claude 3.7 Sonnet perform well in tasks involving complex evidentiary reasoning.
Rather than building its own models, Harvey is opting to fine-tune top-performing existing models for legal-specific applications. The company is also launching a public leaderboard to track the legal performance of various AI models, including detailed research and expert legal insights, rather than relying on simple aggregate scores.
Harvey’s move to integrate a range of AI models—while still valuing its partnership with OpenAI—demonstrates a strategic pivot toward greater flexibility and specialization. “We are incredibly fortunate to have OpenAI as an investor and key collaborator,” said CEO Winston Weinberg. “At the same time, we’re excited to expand options for our customers and push the boundaries of legal AI.