Anthropic, Google Score Win by getting Harvey with OpenAI-backed Harvey | Techcrunch

Popular Legal AI tool Harvey will now use leading foundation models from Anthropic and Google, go beyond strictly using OpenAi’s, Harvey announced In a blog post on Tuesday.

This is remarkable because Harvey is one of the most successful portfolio companies of the OpenAI Startup Fund. The OpenAI Startup Fund is an OpenAI-Associed Fund to support companies that develop products on top of AI technologies, mainly OpenAi’s own. Although Harvey says it does not abandon OpenAi and only adds more models and clouds, this is still a huge coup for the big competitors of OpenAi.

Harvey is one of the first four startups to supported the OpenAI Startup Fund, said it In December 2022. That was back when OpenAI CEO Sam Altman still led the fund. (The others in that first cohort are description, mem and speak.)

Harvey, who has since grown as crazy, is now a startup of $ 3 billion value, said in February, when it announced A series D of $ 300 million under the leadership of Sequoia, with other big names such as coatue, smaller Perkins and the OpenAI Fund stack.

Interestingly, Google’s Venture Arm, GV, Harvey’s Series C of $ 100 million C led in July 2024 (and the OpenAI Fund also participated in that round). But Harvey did not immediately take over the AI ​​models from Google after it had placed the Google’s business company on its CAP table. (GV also participated in Harvey’s Series D.)

So what did Harvey convince now to go beyond OpenAi’s models? The internally developed benchmark of the startup, dubbed Biglaw, showed that a wide range of foundation models is becoming increasingly skilled in a series of legal tasks and some are better in specific tasks than others.

Instead of spending his efforts on training models, Harvey thought, it could easily embrace it, well -performing, reasoning foundation models from other suppliers (eg Google and anthropic via the Cloud of Amazon) and then refining them for the legal market.

The use of different models will also help, because Harvey AI agents creates, says the company.

“In less than a year, seven models (including three non-OAI models) now perform better than the originally benchmarked harvey system on Biglaw Bench,” Harvey wrote in the blog post.

Harvey’s benchmark also showed that different foundation models are better in specific legal tasks than others. For example, it says that Google’s Gemini 2.5 Pro “excels” in legal drafts, but “struggles” with pre-trial tasks such as writing oral arguments because the model does not fully understand “complex evidence rules as heard say.”

OpenAi’s O3 does such a good job before the trial tasks, according to Harvey’s tests, with Anthropic’s Claude 3.7 Sonnet follows on the neighborhood.

In his blog post, Harvey says that it will now also participate in the growing ranks of those who share a public leaderboard with model benchmark performance. The board will rank how big reasoning models do with legal tasks. And the company will not only cook the rankings on one song, but will also publish research in which “top lawyers offer nuanced insights into model performance that is not recorded by single-score benchmarks.”

So, not only by OpenAI-Striden Harvey who takes on the models of competitors, it is also the pressure on his backers (including Google) to continue to prove themselves. Not that OpenAi should be a lot worried about that score. Although AI -Benchmarking is becoming increasingly complex and somewhat political, this is a world where OpenAi is still shining.

“We are incredible happiness that we have OpenAi as an investor in Harvey and important employee in our product,” Harvey CEO Winston Weinberg Techcrunch said in a statement. “And we are energetic to add to our options for customers while we continue to serve the needs of our customers worldwide.”