The popular legal AI tool Harvey will now use Anthropic and Google’s leading foundation model, moving hard to use Openi’s Harvey’s Open AI. Announced In a blog post on Tuesday.
This is noteworthy because the Harvey is one of the initial -backed portfolio companies of the Open AI Startup Fund. The Open AI Startup Fund is a fund -related fund, which companies have the backing of products manufacturing companies in the upper part of AI technologies, mainly by Openi’s own. Although Harvey says he is not abandoning the open, simply adding more models and clouds, it is still a great rebellion for Open’s big rivals.
He said Harvey is one of the first four startups that supported the Openi Startup Fund In December 2022. It came back when Open CEO Sam Ultman was still running the fund. (Other people involved in this first group include explanation, meme and speakers.)
Harvey, who has grown like crazy since then, is now the start of $ 3 billion, she said in February, when it Announced Million, headed by Sevia, piles 300 million series D, other big names such as coat, cleaner Perkins, and an open fund.
Interestingly, Google’s Venture Arm, GV, led Harvey’s 100 million Series C in July 2024 (and the Open AI Fund also participated in this period). But Harvey adopted Google’s AI models immediately after putting Google’s Corporate Venture Firm on his cap table. (GV also participated in Harvey’s Series D.)
So, what did Harvey now agree to go beyond the openings of the open? Internal benchmarks of the Startup, named after Baglavo, showed that different models of the Foundation are growing rapidly in numerous legal works and some are better at specific tasks than others.
Harvey thought that instead of spending his efforts training models, it could easily embrace other shopkeepers (such as Google and Entropic through the cloud of Amazon), the reasoning of the reasoning can easily embrace the argument models and then improve the legal market.
The company says the formation of Harvey AI agents will also help use different models.
Harvey wrote in a blog post, “In less than a year, seven models (including three non -OAI models) now actually improve the benchmark Harvey system on the bench.”
Harvey’s benchmark also showed that different models of the Foundation are better in specific legal activities than others. For example, it states that Google’s Gemini 2.5 Pro is “Excel” on the “legal draft” but “struggles” with trials such as verbal arguments, because the model does not fully understand the complex clear rules such as trial.
According to Harvey’s test, Openi’s O3 performs such trial works well, after which the Anthropic Claude is close to 3.7 Sonit.
In his blog post, Harvey says he will now be included in the growing ranks of the Public Leader of the Model Benchmark Performance. Its board will rank how the model is doing great reasoning on legal activities. And the company will not just rank in one number, but will also publish research where “high lawyers provide proportional insights in the performance of the model that is not taken by a single score benchmark.”
Therefore, not only is the Open AI -led Harvey rival models to adopt, but it is also pressing on its backers (including Google) to prove themselves. It’s not that the open should worry more about this score. Although AI Benchmarking is rapidly complicated and somewhat political, it is the world where the openness is still shining.
Harvey CEO Winston Veneburg told Tech Crunch in a statement, “We are fortunate to get an investor in Harvey and get an open as a key partner in our products.” “And, we encourage users to increase our options as we continue to meet our customers’ needs globally.”