Harvey Token Economics: $20K Per Contract Review — The Cost Architecture Problem Is Now Urgent

The Flank AI briefing for July 10 published Harvey's token economics: the platform's token bill grew 14x in six months, and a single contract review at the top-tier model level can now cost $20,000 in compute. Harvey is understood to be developing a new pricing model. Co-founder Gabe Pereyra's comment on passing token costs to clients as a reimbursable expense — analogous to Westlaw or e-discovery cost pass-throughs — is the most concrete public signal of where Harvey's pricing architecture may evolve. For legal tech operators building on top of frontier models, the $20K-per-task data point is a forcing function for pricing model design: flat seat-based SaaS pricing is commercially untenable when compute costs vary 100x by task complexity and model selection. The LAB-AA benchmark's cost data reinforces the urgency — GLM-5.2 achieves near-equivalent performance to Claude Opus 4.8 at 15% of the cost, and the cost-per-task spread across 28 models is approximately 950x. The operator question is not which model is best but which pricing structure survives the model performance-cost frontier.