Canonical pricing guide · updated September 3, 2026

Legal AI Pricing Models: Compare the Cost of Completed Work

Seat price and token price are not comparable units. A legal team needs one formula that turns every quote into the cost of completed, reviewed work.

Direct answer

Calculate monthly legal AI cost as committed platform fees plus seats or usage, tools, retrieval, failed runs, amortized implementation and qualified reviewer time. Divide that total by accepted completed tasks, not generated drafts. Keep every input, source date and assumption visible so procurement can replace estimates with quote evidence.

Cost stack separating vendor commitment, usage, tools, retries and qualified reviewer time
Decision map. List price is one input. The decision unit is a completed, reviewed task.

Five price models that need different evidence

ModelTypical unitEvidence to request
Seat subscriptionUser per month or yearMinimum seats, included features, usage caps, renewal and offboarding.
Platform commitmentAnnual or monthly minimumCredit consumption, overage, unused balance and term.
UsageTokens, calls, pages or documentsInput, output, cache, tool and batch rates for the exact model.
Transaction or matterCompleted event, document or matterDefinition of completion, exclusions, rework and failure billing.
Sales-led packageCustom quoteSKU, implementation, support, environments, data terms and price-change rule.

Use one transparent formula

Monthly cost = platform commitment + seat cost + model usage + tools and retrieval + failed runs + implementation allocation + reviewer hours × loaded hourly rate.

Cost per accepted task = monthly cost ÷ tasks that passed the agreed review standard.

The formula is intentionally plain. Add a row only when the workflow creates a real cost. Do not hide implementation inside “strategic value,” and do not count a generated response as an accepted task.

Rate cards are inputs, not a vendor ranking

Dated public examplePublished unitWhat it does not include
Claude Fable 5.1$10 input and $50 output per million tokens; $0.25 cache read, checked September 3, 2026.Tools, retries, cloud markup, support and reviewer time.
Gemini 3.8 Flash$0.75 input and $3.75 output per million tokens during the introductory period through December 31, 2026.Grounding, state, surrounding platform and reviewer time.
GPT-5.6 Luna$0.20 input and $1.20 output per million tokens, checked September 3, 2026.Cached-input rules, tools, storage, retries and reviewer time.

These are different model routes, not equivalent legal products. Recheck the linked catalog before using any rate and preserve the quote or order form that actually governs the buyer.

Worked example with explicit assumptions

This illustrative scenario is not a vendor quote: $2,000 platform commitment + $600 model usage + $300 retrieval and tools + 20 reviewer hours at a $150 loaded rate = $5,900 for the month. If 100 tasks pass the defined acceptance standard, completed-task cost is $59. If only 70 pass, the same spend is $84.29 per accepted task. The denominator changes the decision.

Record ranges when evidence is weak. For example, compare 10, 20 and 35 review hours instead of pretending that one early estimate is precise.

Minimum quote record

  1. Product, model, route, region and contract term.
  2. Committed amount, seats, included usage and overage.
  3. Implementation, integration, support and training.
  4. Model, cache, retrieval, storage and tool rates.
  5. Expected volume, retries and failed-run treatment.
  6. Qualified reviewer role, loaded rate and measured minutes.
  7. Accepted-task definition, stop conditions and renewal date.

Keep savings claims bounded

A pilot can show that one workflow used fewer reviewer minutes or cost less during a stated period. It cannot establish universal savings, revenue or firm-wide productivity. Report baseline, sample, exclusions and the worst material failure. When negotiated terms or observed volume are unavailable, keep the field Unknown.

FAQ

What is the best way to compare legal AI prices?

Normalize every option to the cost of an accepted completed task, including commitments, usage, tools, retries, implementation and qualified reviewer time.

Is the cheapest token model the cheapest legal workflow?

Not necessarily. A low generation rate can be outweighed by retries, tools, integration and additional legal review.

Can a firm use public API prices as its budget?

Only as dated inputs. The purchased product, cloud route, order form, usage pattern and reviewer burden can change the actual cost.

Sources checked

Operational information, not legal advice. Verify current terms, account configuration and applicable professional duties before use.

Decision frame · source-linked

Calculate monthly legal AI cost as committed platform fees plus seats or usage, tools, retrieval, failed runs, amortized implementation and qualified reviewer time.

Why now. A compact, source-linked frame makes the next decision portable without replacing the full analysis.

  • Committed platform cost
  • Seat or usage units
  • Tool and retrieval cost

Take this frame with you

Share the decision, not a generic object. The full sources and limits stay on the canonical page.