Legal AI Tool Harvey Expands to Anthropic and Google Models, Boosting Versatility in Legal Tasks

Anthropic and Google have scored a significant victory by securing the use of their advanced AI models by Harvey, a prominent legal AI tool. This news was revealed in a blog post by Harvey on Tuesday, marking a notable shift given the company's deep ties to OpenAI, especially through the OpenAI Startup Fund, which provided early backing to Harvey. Harvey, one of the earliest and most successful investments of the OpenAI Startup Fund, was initially supported alongside other promising startups like Descript, Mem, and Speak. Since receiving its initial investment, Harvey has experienced substantial growth, reaching a valuation of $3 billion in February, following a $300 million Series D funding round led by Sequoia. Other notable investors include Coatue, Kleiner Perkins, and, of course, the OpenAI Fund. Google's venture arm, GV, played a crucial role in Harvey’s development by leading its $100 million Series C funding in July 2024. Despite this financial relationship, Harvey did not immediately integrate Google’s AI models into its platform, maintaining a focus on OpenAI’s offerings. However, recent developments have changed this approach. Harvey’s decision to broaden its AI model portfolio stems from internal benchmarking data. The company developed a comprehensive evaluation system called BigLaw to assess various foundation models on their performance in legal tasks. These benchmarks revealed that a diverse array of models, including those from Anthropic and Google, are becoming increasingly capable in different areas of legal work. For example, Google’s Gemini 2.5 Pro excelled in legal drafting but struggled with pre-trial tasks such as writing oral arguments, particularly due to its limited understanding of complex evidentiary rules like hearsay. In contrast, OpenAI’s O3 model performed well in pre-trial tasks, closely followed by Anthropic’s Claude 3.7 Sonnet. By integrating these varied models, Harvey can leverage each one’s strengths to offer a more comprehensive and effective service to its clients. According to the blog post, Harvey’s new strategy includes adopting multiple high-performing foundation models and fine-tuning them specifically for the legal market. This approach allows the company to focus on enhancing the usability and relevance of AI in legal contexts rather than investing heavily in training from scratch. Moreover, the integration will aid Harvey in developing AI agents that can handle a wide range of legal tasks. To further transparency and competition among AI models, Harvey will now publish a public leaderboard detailing the performance of major reasoning models on legal tasks. Unlike conventional benchmarks that rely on a single score, Harvey’s leaderboard will provide detailed insights, including nuanced feedback from top lawyers on model performance. This move not only diversifies Harvey’s AI capabilities but also increases the scrutiny and accountability of all AI providers, including its former exclusive partner, OpenAI. Despite this shift, Harvey remains appreciative of its partnership with OpenAI. CEO Winston Weinberg stated, “We are incredibly fortunate to have OpenAI as an investor in Harvey and a key collaborator in our product. Adding more options for our customers aligns with our mission to serve global legal needs effectively and robustly.” While the AI benchmarking landscape is becoming more complex and sometimes politically charged, OpenAI continues to excel. Nonetheless, Harvey’s decision to adopt models from its rivals puts additional pressure on OpenAI to maintain its leading edge in AI technologies. This strategic move by Harvey underscores the rapid evolution and competitive nature of the AI industry. By embracing a multi-vendor model, the company ensures it stays at the forefront of technological advancements and continues to meet the sophisticated demands of its legal clientele.