Harvey’s next move: building its own legal AI model

Legal AI company Harvey is developing its own legal foundation models aimed at helping law firms build AI systems trained on their institutional knowledge.

In a post on X, co-founder and president Gabe Pereyra said the initiative has two primary goals: delivering frontier-level intelligence across Harvey's products at lower cost and with stronger security controls, and creating the foundations for firms to build and own their own specialised legal AI models.

The initiative builds on Harvey's existing multi-model approach. The company already gives users access to a range of third-party large language models, and it is understood that its own models will sit alongside those, with different models being deployed depending on the task.

Senior associate model

Harvey’s own models are intended 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 has also open sourced benchmarks designed to evaluate legal work typically performed by associates and in-house counsel. The company said it is expanding those evaluation datasets through a combination of synthetic and human-generated data, while also creating private evaluation frameworks for law firms.

Pereyra said early research conducted with partners including Baseten, Fireworks AI, Applied Compute, Trajectory Labs and Nvidia had shown that post-trained open-weight models could approach the performance of frontier systems on legal tasks.

‘Legal super intelligence’

The announcement comes as legal technology providers increasingly look to differentiate through proprietary models and client-specific AI capabilities as the frontier model providers increasingly push into legal.

Earlier this year Thomson Reuters unveiled its own legal-focused foundation model strategy, building on the company's extensive legal and regulatory datasets.

Kirkland recently partnered with data intelligence giant Palantir to develop AI software that draws on the firm’s own institutional knowledge and workflows, rather than rely on off-the-shelf tools.

Pereyra said Harvey's long-term vision is for law firms, corporations and governments to each own specialised versions of legal AI tailored to their particular expertise and requirements.

"We believe that the best version of legal super intelligence is one where each law firm, enterprise and government owns their own specialised version," he wrote.