Harvey has introduced the Legal Agent Benchmark (LAB) to measure AI agent performance on real-world legal tasks, such as contract review, comparing it to coding benchmarks like SWE-bench. The company advocates for a multi-model strategy in law firms to manage costs and risks, as different models excel at different tasks. Harvey is shifting from a chat-based co-pilot to cloud agents, leading to a dramatic increase in query costs, ranging from $20 to $20,000 due to higher token usage. Gabe Pereyra warns that token-based pricing will become expensive for customers, creating misaligned incentives, but believes market competition will correct pricing. The LAB benchmark, created using agent-led data generation with lawyer review, helps identify cost, latency, and practice area improvements, with predictions that it will be saturated within a year.