- Harvey, an OpenAI-backed legal AI startup, will now use Anthropic and Google models alongside OpenAI’s, citing performance gains.
- Internal benchmarks showed different models excel at different legal tasks, prompting a multi-model approach.
- Harvey will publish a public leaderboard on model performance, increasing transparency and competitive pressure among AI vendors.
Harvey, the fast-growing legal AI tool valued at $3 billion and one of the earliest portfolio companies of the OpenAI Startup Fund, has announced it will now use foundation models from Anthropic and Google in addition to OpenAI’s.
The move is notable given Harvey’s deep ties to OpenAI, which has participated in all of Harvey’s major funding rounds, including its recent $300 million Series D.
While Harvey emphasises it is not abandoning OpenAI, the company’s internal benchmarking, dubbed BigLaw, demonstrated that different foundation models excel at different legal tasks.
For example, Google’s Gemini 2.5 Pro outperforms in legal drafting, while OpenAI’s o3 and Anthropic’s Claude 3.7 Sonnet lead in pre-trial work and reasoning.
By integrating multiple models, Harvey aims to offer its legal customers the best performance across a range of tasks, rather than relying on a single provider.
Harvey’s shift also reflects a broader trend in the AI industry, where startups are increasingly adopting a multi-model approach to maximise accuracy and flexibility.
The company will now publish a public leaderboard to transparently rank model performance in legal applications, adding pressure on AI vendors-including its investors continually improve.
Despite the competitive shake-up, Harvey’s leadership reaffirmed its strong relationship with OpenAI, stating the partnership remains central as the company continues to serve global legal clients.
Edited by Annette George
