Harvey

Generative-AI workflow platform for elite law firms — the OpenAI-backed BigLaw front-runner with $190M+ ARR and an $11B valuation.

1. Core Product / Service

Harvey sells an LLM-powered "co-pilot for lawyers" that sits on top of frontier models (GPT class, with deep OpenAI partnership) and adds legal-domain training, retrieval over a firm's own document corpus, citation, and workflow templates [1][2]. The product surface evolved through three layers:

• Assistant — chat-style legal Q&A grounded in the firm's matters and external case law.
• Vault — secure document workspace where M&A diligence, contract review, and litigation analysis are run as repeatable workflows.
• Agents (2026) — multi-step legal agents handling drafting, redlining, diligence, and research across the entire matter lifecycle. The Mar 2026 funding deck explicitly framed Harvey as moving from "co-pilot" to "agent platform" for law firms and in-house teams [2].

Frontier model access is treated as a strategic moat: Harvey was the first OpenAI Startup Fund portfolio company and reportedly gets pre-release access to OpenAI's legal-tuned variants [3].

2. Target Users & Pain Points

• Buyer: Managing partners / Innovation/IT leadership at AmLaw 200, Magic Circle, and large in-house legal departments.
• Pain points: Junior associate hours on diligence/drafting are the highest-cost line items in BigLaw P&L. Harvey replaces or compresses those hours without forcing partners to sign off on a "ChatGPT in legal practice" risk story — Harvey carries SOC2 / data-isolation guarantees that ChatGPT consumer doesn't.
• Reference customers: Allen & Overy (now A&O Shearman), PwC, Latham & Watkins, ~50 of the top US law firms; >1,000 global clients by Jan 2026 [3][6].

3. Competitive Landscape

Harvey's structural moat is distribution + brand inside BigLaw plus the OpenAI partnership halo; the underlying LLM tech is largely OpenAI-supplied, so the long-term defense rests on workflows, datasets, and partner-level relationships built before competitors scale.

4. Unique Observations

• Pricing model: Per-seat enterprise SaaS. ~$1,000-$1,200 per lawyer / month, 12-month commits, ~20-seat minimums; large firms pay $250K-$5M+/yr depending on headcount [4][7]. Pure seat model — not outcome-based like crescendo / Sierra in CX.
• Implied $/1M tokens consumed: A typical BigLaw user runs heavy retrieval + long context (matter docs, cases). Estimate ~3-5M tokens/lawyer/month at the model layer (Vault diligence runs are 100k-token contexts). At $1,200/seat that implies the user pays roughly $240-$400 per 1M tokens (gross). Underlying GPT-class API cost (2026, with discounts) is ~$2-$5/1M output, so retail-to-cost markup is roughly 50-100×. Most of that gap is workflow + data security + sales motion, not model inference.
• Moat type: Distribution + workflow. Frontier-model access is a temporary advantage; the durable moat is the AmLaw partner Rolodex Harvey built before any competitor scaled into law firms. Regulatory/data-residency tooling reinforces this in EU (Magic Circle).
• Customer profile: BigLaw and large in-house. Explicitly not SMB — the 20-seat minimum prices out solos and boutiques, leaving that segment to Spellbook and CoCounsel.
• Markup multiple over raw API: 50-100× as estimated above. The Mar 2026 Sequoia/GIC deck framed this gap as defensible because the buyer isn't comparing $/token — they're comparing $/associate-hour saved ($400-$800 billable). At that frame, even a 100× markup over OpenAI's raw API still saves the firm money [3][7].
• Revenue trajectory: ~$0 → $50M ARR (mid-2024) → $100M (Aug 2025) → $190M (Jan 2026). Roughly 3× YoY at an absolute scale that few B2B SaaS reach in 36 months [4][2].

5. Financials / Funding

• Total raised: >$800M cumulative; latest round $200M at $11B valuation in March 2026, co-led by GIC and Sequoia [1][2].
• Prior round: $300M at $5B valuation (Aug 2024); $100M at $1.5B (Dec 2023); seed funded by OpenAI Startup Fund (2022).
• ARR: $190M (Jan 2026), up from $100M (Aug 2025).
• Customers: 1,000+ global; >50 of top US law firms.
• Valuation multiple: ~58× ARR — premium reflects strategic positioning + OpenAI tie + BigLaw distribution.

6. People & Relationships

• Co-founders:
  • Winston Weinberg (CEO) — ex-O'Melveny & Myers attorney; the legal-domain anchor.
  • Gabriel Pereyra — ex-DeepMind / Meta AI research scientist; the ML anchor.
• Investors: OpenAI Startup Fund (seed), Sequoia, GIC, Kleiner Perkins, Elad Gil, OpenAI, Google Ventures, Conviction.
• Strategic partners: OpenAI (model + GTM), PwC (channel), Allen & Overy / A&O Shearman.
• Related wiki: robin-ai, spellbook (legal AI peers); crescendo (different vertical, similar "vertical-AI app at Series-late" thesis).