Harvey AI runs Claude Opus 4.6 as its underlying model. Claude Opus 4.6 is available directly through Claude.com — Pro plan, approximately $20/month. The model that scored 90.2% on Harvey's legal reasoning benchmark is accessible to your firm today, without an enterprise contract, without an IT team, and without Harvey's minimum seat requirements.

What BigLaw is paying for is Harvey's platform layer — the case management integrations, the BigLaw-grade security infrastructure, the dedicated support. That platform layer has value for firms with 200+ attorneys and specific enterprise requirements. For a 5–20 attorney firm doing standard contract review, research support, and litigation prep, the question is different: do you need the platform, or do you just need the model?

---

What the 90.2% Score Actually Means for Your Work

Harvey's BigLaw Bench is not an abstract computer science benchmark. It tests the tasks that matter in legal practice: clause identification, legal reasoning under ambiguity, factual synthesis across large document sets, and research memo generation with source grounding.

90.2% on that benchmark means the model is handling those tasks correctly at a rate that would represent meaningful time savings in a small firm context — not because it replaces attorney judgment, but because it handles the first-pass work that currently takes hours.

Three specific capabilities from the benchmark results that translate directly to small firm workflows:

Contract review at volume. Harvey noted strong performance on identifying non-standard provisions and flagging deviations from standard templates. For a small firm reviewing a volume of vendor agreements, NDAs, or client engagement letters, this is the 80% of the work — identifying what needs attorney attention — that currently requires the attorney to read the full document before deciding where to focus. AI-assisted first-pass review changes that: the model flags the clauses, the attorney reviews the flags.

Research memos with citations. 120+ inline citations in a single complex research query is a specific performance claim. For a litigation firm, the time cost of a well-cited research memo — identifying authorities, verifying relevance, formatting citations — is real and significant. Claude Opus 4.6's performance on this task (via the Harvey integration, and available directly) means a first-draft research memo with citations is a starting point, not an end product that gets delivered to a client without review. The attorney reviews the logic and validates the citations. The model does the compilation.

Long document analysis. Claude Sonnet 4.6 (launched in the same period as Opus 4.6's Harvey integration) includes a 1-million-token context window in beta. At roughly 750,000 words per million tokens, that covers a full discovery set, a complete due diligence file, or multiple years of contract history in a single session. The practical change: you're no longer manually chunking a 400-page document into sections and losing context between sessions.

---

Which Interface Do You Actually Need?

Harvey is not the only way to access Claude Opus 4.6's legal reasoning capability. The three realistic options for a small law firm:

Direct Claude (Pro or Max plan): $20–$100/month per user. Opus 4.6 available on both tiers. No legal-specific workflow layer — you work directly with the model. Best for: attorneys who want to run structured prompts for contract review, research support, and document analysis without a platform built around it. The trade-off: you bring the workflow structure, the model brings the reasoning.

Clio Duo: Clio's AI layer integrated with Clio's case management system. Uses Claude under the hood for some functions. Best for: firms already running on Clio who want AI capabilities embedded in their existing workflow without a separate tool. The Clio integration means matter context is available inside the AI conversation — you don't have to paste case details each time.

Harvey AI: Enterprise contract, minimum seats, designed for AmLaw 200 and global firm use cases. Best for: large firms with specific infrastructure requirements, BigLaw integrations, and dedicated legal AI support needs. Not the right choice for a 10-attorney firm paying for platform capabilities they won't use.

For most small law firms evaluating this decision: start with direct Claude access. If you want workflow integration with existing case management, evaluate Clio Duo. Reach for Harvey when your firm's volume and infrastructure requirements make the enterprise platform layer worth the cost.

---

The Dismissal Worth Pushing Back On

The most common small firm response to AI legal tools is: "That's for large firms with big AI budgets." Harvey reinforces that perception — it targets Am Law 200 firms and above, and its marketing reflects that.

But the underlying model doesn't know which firm it's working for. The 90.2% benchmark score reflects Claude Opus 4.6's legal reasoning capability — which is available at direct Claude pricing. The gap between what Harvey does for a 200-attorney M&A group and what direct Claude does for a 10-attorney general practice is a gap in workflow integration, not in the model's ability to review a contract or draft a research memo.

The strategic question this raises for a small firm isn't "should we use Harvey?" — it's "which of these three legal tasks are we doing repeatedly, and is AI assistance worth the time to test?"

If you're reviewing NDAs weekly, the time savings from structured AI review are measurable. If you're drafting research memos with citations monthly, the time savings are measurable. If you're doing neither, the benchmark score doesn't matter.

The Thomson Reuters 2026 AI in Professional Services Report found that firms with a written AI strategy — a documented decision about which tools they use and for what tasks — are three times as likely to achieve positive ROI from AI investment as firms without one. The written strategy is not a long document. It's a decision: here are the three things we're testing this quarter, here's who owns the pilot, here's what we're measuring.

The Harvey + Claude Opus 4.6 benchmark is a data point that answers "is this model good enough for legal work?" at 90.2%. The more useful question for your firm is: "have we decided what we're going to use it for?"

---

Where to Start This Week

If you're a 5–20 attorney firm that has been following AI legal tool news from a distance:

1. Open a Claude Pro account ($20/month at claude.ai). Use Opus 4.6 for one document review task this week — a contract you're already reading, an NDA coming in for a new client, or a brief excerpt from a matter in progress.

2. Use a structured prompt: "Review this agreement and identify: (a) non-standard indemnification provisions, (b) missing standard protections, (c) ambiguous definitions, and (d) unusual termination triggers. For each issue, explain the risk and suggest alternative language." Compare the AI's flags to your own review.

3. Track for 30 days. What did it catch that you would have caught anyway? What did it catch faster? What did it miss? After 30 days, you have data — not a hypothesis — about whether AI-assisted contract review saves time at your firm.

That's the written strategy for contract review. One tool. One task. One month. The benchmark score at 90.2% is useful context. Your 30-day tracking is the evidence that matters.

(Sources: Harvey AI blog; MLQ.ai; Anthropic; DreamLegal; Thomson Reuters 2026 AI in Professional Services Report)