Contract Review Is Where AI Earned Its Place in Law Firms
Of all the AI use cases in legal, contract review is the one with the most settled ROI. First-pass review of an NDA, a vendor agreement, a credit-line covenant — every firm has lawyers doing this work at a billing rate that's an order of magnitude above the underlying compute cost.
The specialty AI vendors that own this category — Harvey, Thomson Reuters Co:Counsel, Spellbook, Ironclad AI, LinkSquares — have priced accordingly. Per-lawyer fees of $200–500/month or per-document fees of $1–5/contract are common. Either pricing shape captures the value of the lawyer-hour replaced, not the cost of producing the review.
The cost of producing the review is much smaller. Showing the math is the post.
What a Contract Review Actually Costs Per Token
A typical first-pass review is about 3,000 input tokens (the contract + checklist of clauses to flag) and 800 output tokens (the structured summary with risk classifications and cited clauses). Cost-per-contract on the major models:
| Model | Input ($/MTok) | Output ($/MTok) | $ per contract | When to use it |
|---|---|---|---|---|
| Claude Opus 4.7 | $15 | $75 | $0.105 | High-stakes M&A, complex covenants |
| GPT-5 | $10 | $30 | $0.054 | Mixed-complexity transactional work |
| Claude Sonnet 4.6 | $3 | $15 | $0.021 | Standard contract review workhorse |
| Gemini 3 Pro | $3.50 | $10.50 | $0.019 | Long-context (full deal-room) reviews |
| Claude Haiku 4.5 | $1 | $5 | $0.007 | High-volume NDA sweeps, intake triage |
| Llama 4 / DeepSeek-R1 (self-hosted) | ~$0 | ~$0 | ~$0 | Inside the firm's network |
The most expensive frontier model reviews a contract for 10 cents. The mid-tier workhorse: 2 cents. The cheap option: less than a cent. Self-hosted: marginal cost is electricity.
Monthly Bills at Three Scale Tiers
- Solo / small firm (10 lawyers): ~500 contracts/month
- Mid-market (75 lawyers): ~5,000 contracts/month
- AmLaw 100 (200 lawyers, active M&A practice): ~30,000 contracts/month
Monthly cost using Claude Sonnet 4.6 vs the per-document and per-lawyer alternatives:
| Approach | Pricing shape | Solo (500/mo) | Mid-market (5K/mo) | AmLaw (30K/mo) |
|---|---|---|---|---|
| Harvey | ~$400/lawyer/mo | $4,000 | $30,000 | $80,000 |
| Thomson Reuters Co:Counsel | ~$300/lawyer/mo | $3,000 | $22,500 | $60,000 |
| Spellbook / Ironclad AI / LinkSquares | ~$2/contract or ~$100/lawyer | ~$1,000 | ~$10,000 | ~$60,000 |
| Direct API — Claude Sonnet 4.6 | Token-based | ~$11 | ~$105 | ~$630 |
| Direct API — GPT-5 | Token-based | ~$27 | ~$270 | ~$1,620 |
| ibl.ai self-hosted (Llama 4 / DeepSeek-R1) | Flat license + GPU | ~$1,500 | ~$3,000–5,000 | ~$5,000–8,000 |
At AmLaw scale, Harvey is ~130× more expensive than the same contracts reviewed on direct Sonnet API, and ~12× more expensive than the all-in self-hosted line.
The Per-Document and Per-Lawyer Traps
Per-document pricing ($1–5/contract) feels aligned to value. It isn't. A vendor's marginal cost is fractions of a cent; the $2 fee is value capture. The math doesn't get better with volume because the price doesn't drop.
Per-lawyer pricing ($200–500/month) is worse. The firm pays for paralegals, document-review attorneys, and staff who use the tool occasionally at the same rate as the partner running a $50M M&A deal. The vendor's "per-lawyer" billing is really "per-license-seat-counted," which is to say, "per the firm's headcount."
The structural problem with both: privileged work product is sitting in a third party's cloud. ABA Model Rule 1.6 puts the obligation to make "reasonable efforts to prevent the inadvertent or unauthorized disclosure of" client information on the lawyer — not the vendor. Several state bars are now treating that as incompatible with sending privileged documents to a managed AI cloud, regardless of the DPA.
Why Self-Hosting Is the Privilege-Compatible Answer
The privilege analysis collapses if the model runs inside the firm's network. There is no third-party custodian. There is no subpoena reach to the vendor for the firm's working drafts. There is no DPA refresh every time the vendor changes sub-processors.
For contract review specifically, the operational benefits compound:
- The firm's playbook lives in the firm's repo. Standard markups, preferred fallback positions, jurisdiction-specific tweaks — all configured in the agent, all version-controlled by the firm, all updatable the day a partner changes a position.
- Conflicts checking integrates with the firm's existing systems. Connection to iManage / NetDocuments / SharePoint happens inside the firm's network; no document leaves the perimeter to get reviewed.
- Bulk diligence runs use the cheap model. A 5,000-document M&A diligence dataset reviewed on self-hosted Llama 4 has a marginal cost of GPU time, not a $25,000 vendor invoice.
What Stays the Same, What Changes
Self-hosting contract-review AI doesn't mean rebuilding the firm's legal-tech tooling. The matter-scoped workspaces, the chat UI, the citation-checking, the document-management integration, the audit logs, the multi-agent orchestration — all stays managed by ibl.ai. The compute, the model, and the privileged documents move inside the firm's network.
What disappears: the $60–80K/month Harvey bill (or the $30K Co:Counsel bill, or the $10–60K specialty-tool bill).
What appears: a self-hosted contract-review capability the firm owns, with a model-routing recipe each practice group designed:
- Opus for high-stakes M&A redlines, complex covenant negotiations, appellate brief work
- Sonnet for standard transactional review (the bulk)
- Haiku for NDA sweeps and intake triage
- Llama 4 self-hosted for bulk diligence where even pennies per document add up at 30K+ volume
Run the Numbers for Your Firm
For the segment-wide cost-math context (not just contract review), see AI Cost Math for Law Firms: Per-Seat vs Usage-Based in 2026.
For the deployment comparison side-by-side — including ABA Model Rule 1.6 posture, privilege protection, and air-gapped options for the most sensitive matters — see Self-Hosted AI vs ChatGPT Enterprise for Legal.
For the broader policy framework — what a law-firm AI policy should cover and why owned/air-gapped deployment is the control that makes it enforceable — see AI Policies for Law Firms: A Practical 2026 Guide.
For the broader pricing landscape across every model and per-seat vendor, the hub: What Does AI Actually Cost in 2026?.
Why Family-Owned and New York Matters Here
A law firm's AI vendor relationship for a workload as central as contract review is a multi-year commitment that touches privileged client work product. ibl.ai is family-owned and operated from New York, NY — a long-term partner with a perpetual platform license and no investor exit pressure. The runtime is open source. The privileged data stays inside the firm's network. The math works at a 5-lawyer boutique or a 2,000-lawyer global firm.