The Brief:
Harvey is building its own legal AI models, sitting alongside the third-party LLMs it already runs on.
The move comes as frontier players like OpenAI and Anthropic push deeper into legal, forcing legal tech to find new ways to stand out.
Harvey isn’t content just plugging into everyone else’s models anymore.
Harvey’s own model
The legal AI heavyweight has confirmed it’s building its own legal foundation model series, sitting alongside the third-party LLMs that already power its platform.
Co-founder and president Gabe Pereyra dropped the news on X. Pereyra said the model has two goals: “1. Allow us to serve frontier intelligence across our product surface areas at an affordable price and a strong security posture. 2. Create the foundations for law firms to build their own specialized models and own their own intelligence.”
So, what does that actually look like?
It’s less chatbot and more digital senior associate.
The model is built to act “much like a senior associate”, tackling complex client matters that “span months and take dozens of associates,” Pereyra wrote. “The agentic system will learn to control legal tech tools, sub agents and ask for help from frontier models or human partners.”
Harvey is also throwing its doors open. Training data and benchmarks are being open sourced, built off real associate and in-house work. Pereyra said the company plans to “invest heavily in working with research partners and open sourcing our data, models and research as much as possible.”
Working with research and infrastructure partners Baseten, Fireworks AI, Applied Compute, Trajectory Labs and Nvidia, the company says its post-trained open-source models are now performing nearly as well as the frontier models like GPT-5.
Why now?
The timing isn’t a coincidence.
The frontier labs aren’t staying in their lane anymore, they’re coming straight for legal. Last month, Anthropic launched Claude for Legal, its own dedicated suite of AI tools for law firms and in-house teams. Meanwhile, OpenAI has poached Jason Boehmig, founder of CLM pioneer Ironclad, to lead product for its own legal vertical, with plans for a legally focused suite of tools reportedly in the works.
As Anthropic and OpenAI build legal capability straight into their core platforms, legal AI players like Harvey need a harder edge to stay relevant. Owning proprietary models trained on real legal workflows is exactly that.
Others take the same view.
Kirkland & Ellis is spending US$500m to develop its own AI platform, built off the “collective intelligence” of its lawyers. Thomson Reuters is gearing up to launch its own legally-trained LLM, built on open source models and the company’s massive proprietary data store.
Not everyone’s convinced.
Artificial Lawyer thinks most legal work product doesn’t really have a moat — documents get shared, lawyers move firms, precedents leak out. What’s actually hard to replicate is the relationship itself. It’s built on the partner’s style, the GC’s quirks and years of working chemistry. That’s human, and not something any model can fully replicate.
Source: X, Non-Billable, Artificial Lawyer