Claude Fable 5 for Contract Review

Claude Fable 5 is a strong tool for contract review. It is Anthropic's most capable public model for knowledge work and vision, released June 9, 2026, and it reads long agreements, extracts clauses, flags risky terms, and supports redlining well.

This page covers the concrete review workflows Fable 5 fits: clause extraction, risk flagging, redline support, summarizing long agreements, and comparing versions. It also covers the two things a firm must plan around before any privileged contract goes in.

The first is confidentiality. As a Mythos-class model, Fable 5 carries a mandatory 30-day safety retention on business accounts, so full zero-data-retention is not available. The second is human review: a licensed attorney must review every output, because the model is a drafting aid and not legal advice.

Why Claude Fable 5 fits contract review

Claude Fable 5 fits contract review because the task is document comprehension, and that is its strongest suit. It is Anthropic's most capable public model for knowledge work and vision, so it reads long agreements closely and handles scanned or PDF exhibits without losing the thread.

Contract review is interactive and document-in-hand. You give the model a specific agreement, and it works the clauses with you. Fable 5 also sustains longer multi-step tasks than earlier Claude models, which suits reviewing a full contract set in one pass.

Pricing is $10 per million input tokens and $50 per million output tokens. Input is the contract you send in; output is the markup the model writes back. A single agreement costs cents, not dollars, per run.

• Reads long agreements and full contract sets closely.
• Vision handles scanned contracts, exhibits, and PDFs.
• Sustains multi-step review across a whole document.
• Cheap enough per run to review at volume.

Concrete contract-review workflows

Fable 5 supports the core review workflows lawyers actually run. It extracts clauses into a structured list, flags terms that look risky or off-market, and drafts redline language you can accept, reject, or rewrite.

It is also strong on the reading-heavy tasks around review. It summarizes a long agreement into its key obligations, and it compares two versions of a contract to surface what changed between drafts.

Treat each output as a first pass. The model does the mechanical reading fast and consistently; the lawyer makes the substantive call on every flag, edit, and summary.

• Clause extraction: pull defined terms, obligations, and key clauses into a list.
• Risk flagging: surface non-standard, one-sided, or missing terms.
• Redline support: draft proposed edits for a lawyer to approve.
• Summarizing: condense a long agreement to its core obligations.
• Version comparison: show what changed between two drafts.

Confidentiality, privilege, and the 30-day retention rule

This is the specific thing law firms must plan around. As a Mythos-class model, Claude Fable 5 carries a mandatory 30-day safety retention on business accounts. Anthropic uses that data only for safety, then deletes it, but full zero-data-retention is not available on this model.

Contracts are among the most sensitive material a firm handles, and many are privileged or under an NDA. Under ABA Model Rule 1.6, that 30-day window is a confidentiality decision you make before, not after, sending a document. If a client contract or an ethics opinion demands zero retention, this model does not meet it today.

The practical path is to use the API or Enterprise plan under a HIPAA BAA and appropriate terms, and never the consumer tier for client data. Confirm your engagement letters permit a 30-day safety window, or obtain written client consent, before any privileged agreement goes in.

A licensed attorney must review every output

Human review is not optional. Claude Fable 5 is a drafting aid, not a lawyer, and it gives no legal advice. Every clause flag, redline, summary, and version diff it produces is a draft a licensed attorney must review before it is relied upon.

The model can produce confident output that is wrong. It may miss a buried term, misread an off-market clause as standard, or state an obligation incorrectly. On a contract, a missed risk is the whole point of the review, so the attorney owns the substantive judgment on every item.

Under ABA Model Rules 5.1 and 5.3, the supervising lawyer is responsible for the work regardless of how the draft was produced. Build a named-reviewer checkpoint into the workflow so no AI-assisted output leaves the firm unchecked.

Limits and where to be careful

Fable 5 has real limits on contract work beyond the retention rule. It can miss context that lives outside the document, like a side letter, an oral agreement, or the deal's commercial intent. It reviews the text in front of it, not the full matter.

It also blocks high-risk cyber and bio requests and falls back to Claude Opus 4.8 when it detects one. Anthropic reports more than 95% of sessions never trigger this fallback, and normal contract work almost never hits it. Still, check which model answered on any research touching those topics.

For availability, Fable 5 returned to worldwide general availability on July 1, 2026, after a June 12 export-control suspension that the US Commerce Department cleared. If your firm runs it through AWS Bedrock, Google Cloud, or Microsoft Foundry, confirm it is live in your region, since those are re-enabling in stages.