What is Harvey?

Harvey is building frontier legal AI on open-weight models under the Harvey Tenant brand, and enabling law firms to train their own specialized models, signaling a move away from dependence on frontier API providers.

“Over the past six months, Harvey's research agenda has focused on two goals. One, building frontier legal intelligence using openweight models. And two, creating systems to allow law firms to build their own specialized models and their own intelligence.”

Harvey now spends 99% of its AI budget on non-frontier models, with any remaining frontier spend likely used for distillation rather than production inference.

“99% of their spend is coming off the Frontier models.”

Harvey's residual frontier-model spend is best explained as distillation spend, not production inference, meaning frontier models are used to train smaller proprietary models rather than to serve live queries.

“Any spend they do have with the Frontier models is likely, even if it's like 5% of their spend, to be them doing distillation.”

Trajectory.ai compressed model onboarding time from 3 months on its first Harvey engagement, to under 1 month, and then to approximately 1 week for subsequent customers.

“The first engagement, it took like three months to like set up the entire thing. We're kind of building the airplane as it was flying and now we're able to train a model with Harvey in under a month. and then onboarded a new customer and within a week we were able to train a model and really get the flywheel going.”

Harvey and Fireworks pair an open-weight GLM worker model with an Opus advisor for legal tasks, reportedly outperforming Opus alone at a fraction of the cost.

“Harvey and Fireworks pair an open weight GLM worker with an Opus advisor for legal tasks, seeing improved performance over Opus alone for a fraction of the cost.”

Harvey requires both a field deployment engineer and a lawyer on every enterprise deployment, signaling a high-touch, expert-paired model that may define the standard for legal AI rollouts.

“Every deployment they do at Harvey has an FD and a lawyer.”

Harvey used a router combining an open-source model to beat Anthropic Claude Opus 4 on outcome quality at lower cost, a notable enterprise AI routing result.

“They got better outcomes than Opus 4 either 4.7 or 4.8 at a lower cost.”

Harvey (legal AI) forked the open-source Qwen/Kimmy K3 model to build a sovereign legal AI, keeping full control of data and model signal. A notable template for vertical AI sovereignty.

“Harvey has forked I believe it was Quen >> Kimmy K3. >> Kimmy K3 to make their own legal AI sovereign.”

Harvey (legal AI): reportedly now matched by Claude out of the box despite years of legal specialization, signals rapid commoditization of vertical legal AI.

“What I've been hearing lately is that Claude out of the box is just as good as Harvey.”

App-layer companies are forking open-source frontier models to go sovereign, directly countering the assumption that frontier labs will remain the default infrastructure for vertical AI businesses.

“People are starting to realize I just can't trust the Frontier models not to steal my business.”