Harvey
An AI startup focused on the legal industry, providing vertical AI solutions for law firms and corporate legal departments. Its offerings cover professional legal tasks including document review, case research, and contract drafting.
Core Facts
Timeline (last 90 days)
Harvey LAB comes from within the industry rather than an academic institution, and its task design is closer to the real workflows of law firms.
Harvey LAB (Legal Agent Benchmark) simulates Agent tasks from real lawyer workflows, requiring the model to have comprehensive capabilities such as multi-step reasoning, document retrieval, clause comparison, and structured output.
The background of OpShield's creation is that after Guardrails AI was acquired by Harvey, the market lacked an independent, self-hostable security alternative.
In 2024, the Guardrails AI team was acquired by the AI legal tech company Harvey, with core members joining Harvey, and the original project's independent development pace and community maintenance efforts subsequently weakened.
Harvey and van der Hoeven proposed an O(n log n) integer multiplication algorithm, which is considered to be asymptotically near-optimal.
The experimental design was inspired by a workflow from legal AI company Harvey: a small company acquisition review.
Charging by seat while being billed by token on the backend creates a structural cost misalignment, which is essentially a subsidy coupon with an expiration date.
Harvey reduced the cost of one legal task from $52 to $30 (data from its own testing, lacking third-party verification).
After Harvey launched a self-developed model based on China's Kimi K3, its gross margin turned positive, but Kimi K3's unit price is only 40% cheaper.
Harvey's token usage grew 20x within a year, revenue doubled, and gross margin fell from positive 50% at the start of the year to negative 50% in June.
19 more timeline events
All Facts (20)
The API model identifier for GPT-6 Astra is gpt-6-astra.
90%VerifiedHarvey is an AI startup focused on the legal industry.
75%UnverifiedHarvey采用'协作优先'（collaborative by design）的工程哲学，从第一天就把协作当作核心需求来设计，而非先做好个人工具再打通
85%UnverifiedHarvey的核心产品是面向律师事务所和企业法务部门的AI助手，能够处理合同审查、法律研究、诉讼分析等专业任务
85%UnverifiedGPT-6 Astra is capable of generating legal documents with clearer structure and closer alignment to specific contexts.
70%Unverified协作式Agent在实践中遇到的问题包括信息过载、上下文污染和协作流程设计不当导致的冲突处理机制不完善
70%UnverifiedHarvey LAB comes from within the industry rather than an academic institution, and its task design is closer to the real workflows of law firms.
50%UnverifiedHarvey LAB (Legal Agent Benchmark) simulates Agent tasks from real lawyer workflows, requiring the model to have comprehensive capabilities such as multi-step reasoning, document retrieval, clause comparison, and structured output.
50%UnverifiedThe background of OpShield's creation is that after Guardrails AI was acquired by Harvey, the market lacked an independent, self-hostable security alternative.
50%UnverifiedHarvey and van der Hoeven proposed an O(n log n) integer multiplication algorithm, which is considered to be asymptotically near-optimal.
50%UnverifiedThe experimental design was inspired by a workflow from legal AI company Harvey: a small company acquisition review.
50%UnverifiedHarvey reduced the cost of one legal task from $52 to $30 (data from its own testing, lacking third-party verification).
50%UnverifiedHarvey's token usage grew 20x within a year, revenue doubled, and gross margin fell from positive 50% at the start of the year to negative 50% in June.
50%UnverifiedCharging by seat while being billed by token on the backend creates a structural cost misalignment, which is essentially a subsidy coupon with an expiration date.
50%UnverifiedAfter Harvey launched a self-developed model based on China's Kimi K3, its gross margin turned positive, but Kimi K3's unit price is only 40% cheaper.
50%UnverifiedLegal AI products emphasize final attorney review as an unavoidable node in the workflow.
50%UnverifiedHarvey focuses on specific legal workflows such as contract review, legal research, document drafting, and due diligence.
50%UnverifiedPlatforms such as Harvey typically address confidentiality concerns through private deployment, data isolation agreements, and zero data retention clauses.
50%UnverifiedHarvey's core objective in integrating GPT-6 Astra is to transform complex legal context into higher-quality document drafts.
50%UnverifiedHarvey and Engram used the model to build a synthetic law firm, achieving an average score of 67 across 250 legal tasks, higher than all models evaluated in the study.
50%