Interview, Fireside Chat
Harvey Co-Founder Gabe Pereyra on the Token Pricing Reckoning Coming for AI
- Token consumption and associated costs are projected to surge significantly over the next two years, with single queries potentially reaching $20 and complex document reviews $20,000, creating a financial environment where billing may shift from token-based to subscription models due to forecasting complexities.
- Within the next six months, capability constraints in legal AI are expected to transition to cost constraints, necessitating intelligent task routing and driving demand for specialized, smaller models that deliver frontier performance without the expense of trillion-parameter systems.
- The market is predicted to converge on pricing structures that reflect the actual value of work performed, influenced by increasing competition among inference providers, the catch-up of open-source models in performance, and the saturation of legal agent benchmarks within one year.
- Business strategies are anticipated to evolve into parallel models comprising traditional enterprise seats and consumption-based tiers, while the industry develops ecosystems for granular billing visibility to align incentives between providers and customers.
- Organizational intelligence will become the defining focus of the coming year, shifting emphasis from individual agent capabilities to the collaboration between lawyers and human-agent teams within unified platforms.
- Law firms are expected to mitigate conflict risks and manage platform costs by supporting multiple model providers, abstracting underlying infrastructure, and utilizing unique, non-public data to own their systems rather than relying solely on public general-purpose models.
- Adoption inflection points will be driven by models capable of generating entire documents that outperform human workflows, forcing routine changes, while the research landscape may become less transparent regarding specific breakthroughs due to reduced open publishing.
- Open-source communities and internal research teams are expected to collaborate on post-training open-weight models to close performance gaps with closed-source frontier models, potentially spawning new research directions in agent harnesses and legal-specific skills.
- Future value is anticipated to derive from solving complex data organization and optimization challenges within shared spaces between law firms and clients, adhering to ethical and regulatory constraints, while competition drives a trend similar to cloud infrastructure multi-provider strategies.