The popular legal tool AI Harvey will now use the leading models of foundations from Anthropic and Google, going beyond strictly with Openai, Harvey announced In the blog post on Tuesday.
This is noteworthy because Harvey is one of the most successful portfolio corporations of the OPENAI startup fund. The OPENENAI Startup Fund is a fund related to OPENAI for corporations developing products in addition to AI technology, mainly its own OPENNAI. While Harvey says that he does not abandon OpenAi, he simply adds more models and clouds, it is still a huge coup for large Opeli competitors.
Harvey is one of the first 4 startups that the OPENENAI starting fund supported, he said in December 2022. It was when the general director of Opeli, Altman himself, continued to run the fund. (Others in this first cohort include a description, meme and speak.)
Harvey, who has turn out to be crazy since then, is now a startup value $ 3 billion, he said in February when this announced The D Series value $ 300 million run by Sequoia, with other large names, such as Coatue, Kleiner Perkins and the OpenAI fund.
Interestingly, Google Venture ARM, GV, led the Harvey C series in the amount of $ 100 million in July 2024 (and the Openai fund also participated in this round). But Harvey didn’t immediately accept AI Google models after he placed the company of the Google enterprise on the Cap table. (GV also participated in the H Harvey series.)
So what convinced Harvey to go outside the OpenAI models? The benchmark, named Biglaw, internally developed by the startup, showed that many different foundation models are becoming more and more expert in many legal tasks, and some are higher in specific tasks than others.
Harvey thought that as an alternative of spending training models, he could simply accept high -performance, reasoning foundation models from other suppliers (e.g. Google and Antropic via Amazon clouds), and then refine them to the legal market.
The company says that the use of varied models can even help when Harvey creates AI agents.
“In less than a year, seven models (including three models from outside OAI) are now surpassed by the originally compared Harvey system on the Biglaw bench,” Harvey wrote in the blog post.
Harveya benchmark has also shown that various foundation models are higher in specific legal tasks than others. For example, he says that Google Gemini 2.5 Pro “stands out” in legal drawing, but “fights” with pre -trial tasks, such as writing oral arguments, because the model does not understand the fully “complex evidence principles such as rumors.”
O3 Openai performs such pre -trial tasks, in accordance with Harvey’s testing, and Antropic’s Claude 3.7 Sonnet is much like the back.
In his blog on the blog, Harvey claims that now he can even join the growing ranks of individuals dividing the public plaque of the reference model results. Its management argues how the major models of reasoning perform legal tasks. And the company won’t only reduce to the rating to one number, but also publish studies in which “the best lawyers provide detailed information on the performance of the model that is not recorded by sole proprietors’ references.”
Therefore, Harvey, Opennai, accepts competitors’ models, but also increases pressure on his supporters (including Google) to proceed to prove. Not that Opeli should worry about this result. While AI comparative evaluation is becoming more and more complex and a bit political, it is a world in which Opeli is still shining.
“We are amazing luck that we have OpenAI as an investor in Harvey and a key collaborator in our product,” said the general director of Harvey Winston Weinberg in a statement. “We are energetic to increase our options for customers because we still meet the needs of our clients around the world.”