Legal

Opus 4.8 vs GPT-5.5
for legal review

This page compares two current models on one job: reviewing contracts and legal documents. It looks at accuracy, cost and prompting, and it ends with a fair way to test them on your own files, with a lawyer still in the loop.

Jul 17, 2026 · 11 min read

The verdict
Which model wins each job

Claude Opus 4.8 is the safer default when the job is reading a whole agreement, spotting issues and drafting in context. GPT-5.5 is the stronger pick when the job is tightly defined extraction or a narrow revision that must fit a fixed shape.

That split shows up across the public legal benchmarks3, vendor evaluations45 and each provider's own guidance810. On the Legal Benchmarks June 2026 run, Opus 4.8 passed 67.6% of contract-drafting tasks against 41.2% for GPT-5.5, while the two were close on plain extraction at 83.3% and 80.0%.3

The practical move is to match the model to the job, and to keep a lawyer in the loop either way. For issue spotting, contradiction checks and redlines, start with Opus 4.8. For pulling fixed fields into a table or making one narrow edit, start with GPT-5.5. For a mixed workload, use GPT-5.5 to extract, then Opus 4.8 to review and draft. Both still make mistakes that a person has to catch.7

Who this is for
Which legal roles this fits

Start with Opus 4.801

In-house legal teams

You review NDAs, MSAs and policies across the business and care most about spotting issues and missing clauses. Opus 4.8 leads the public contract-review tests, so it is the safer first read.

Use both02

Transactional lawyers

You redline agreements and prepare issues lists under deadline. Let GPT-5.5 pull the deal points into a table, then have Opus 4.8 spot conflicts and draft the revisions.

Start with GPT-5.503

Legal operations

You run high-volume extraction into a database or contract system. GPT-5.5 follows a fixed schema well, as long as you require an evidence span for every field.

Test both04

Legal tech builders

You are wiring a model into a review product. Test on your own jurisdictions and playbooks, since retrieval and orchestration change the result more than the raw model.

What we compared
Review accuracy not the app

This page treats each one as a legal-review model, not as a finished legal product. So it weighs the parts of review that show up in real matters.

Those parts are clause and deal-point extraction, missing-clause detection, playbook comparison, risk spotting, redlining and preparing an issues list. Official sources come first, then independent legal benchmarks and vendor tests with clear methods.

We left tools out of the spec table on purpose. File parsing, retrieval and citation behaviour depend on the app around the model, so the same model can behave very differently in a chat product, in the API or inside a legal platform. Clio's strong GPT-5.5 results, for example, came from its full Vincent system rather than the raw model.11

One timing note. GPT-5.5 has since been superseded, as OpenAI now recommends GPT-5.6 as its flagship and Anthropic has released the pricier Claude Fable 5 while keeping Opus 4.8 current.1314 This comparison still helps teams on a pinned GPT-5.5 deployment or comparing models at about the same price.

Specs at a glance
The review-relevant numbers

The model facts that actually affect a review job. Tool features are left out, since they change with the app around the model.

Spec
Claude Opus 4.8
GPT-5.5
Why it matters
Context window
1,000,000 tokens
1,050,000 tokens
Room to load a long agreement or a document set in one pass
Max output
128,000 tokens
128,000 tokens
How much redline or issues list it can return at once
Input price
$5 per million
$5 per million
Matched, so input cost is a wash
Output price
$25 per million
$30 per million
Opus 4.8 is cheaper on output-heavy redlines and issues lists
Long-context price
Full window at standard rates
Higher rates past 272,000 input tokens
Affects the bill on large document sets
Default posture
Reviews its work and flags weak assumptions
Tight instruction following
Shapes how much the model second-guesses its own draft

Figures from Anthropic and OpenAI documentation, July 2026. These are model specs, not a promise that an app exposes the full context window.

Head to head
Where each model leads by job

The answer changes by subtask, not by brand. This is the main analysis: which model has the edge on each part of a review workflow, and what backs it up.

Job
Better choice
Why the edge exists
Best evidence
Whole-contract review
Claude Opus 4.8
On the Legal Benchmarks June 2026 run, Opus passed 67.6% of contract-drafting tasks against 41.2% for GPT-5.5, including redlines and instructions that conflict with the source.
Legal Benchmarks, June 2026
Clause and deal-point extraction
Close, test both
The same run gave Opus 83.3% and GPT-5.5 80.0% on extraction. GPT has more supporting evidence for schema-first, parsing-heavy workflows.
Legal Benchmarks, June 2026
Following a narrow revision
GPT-5.5
LegalOn's 200-example revision test reported 91.0% instruction adherence and 87.5% all-criteria accuracy, often making just the requested edit.
LegalOn evaluation
Contextual drafting and redlining
Claude Opus 4.8
Opus led the drafting test by a wide margin, and Harvey saw review-and-revise behaviour that improved its drafts.
Legal Benchmarks and Harvey
Flagging uncertainty
Claude Opus 4.8, with care
Anthropic reports fewer unsupported progress claims and more proactive flagging, but Harvey still found Opus sometimes missed ambiguity that needed a person.
Anthropic and Harvey
Avoiding fabricated authority
GPT-5.5, neither safe
HAQQ rated GPT-5.5 highest for accuracy at 8.41 out of 10 with a 3% made-up-citation rate, yet it still invented at least one citation.
HAQQ, June 2026
Binary playbook compliance
Claude Opus 4.8, test both
LegalOn found GPT-5.5 produced more false positives than GPT-5.4 on a 494-decision review and sometimes missed absent language.
LegalOn evaluation
API cost
Claude Opus 4.8
Input prices match, but Opus output is $25 against $30, and GPT-5.5 charges more past 272,000 input tokens.
Official pricing

Better-choice calls come from official docs, pricing and independent legal benchmarks cited at the end of the page. Where the evidence is close or indirect, the row says so.

How to test
A fair test on your own matters

A useful test feels boring. Same prompt, same files, same conditions, same scoring. Then judge what your team actually pays for: did it find the right clauses, flag what was missing or ambiguous, avoid inventing text, and need less lawyer editing.

Sample01

Pick three to five matters

Use real files: a standard NDA, an MSA with hidden conflicts, a poor scan and a multi-document amendment chain. Toy contracts do not show how a model behaves on your work.

Prompt02

Give both the same prompt

One prompt that sets the task, the jurisdiction, the required fields and the output shape. Neither model gets a richer version. If you change the prompt mid-test, apply the change to both.

Setup03

Use the same setup

Same files and the same place to run them, whether that is the API, a chat product or a legal platform. Orchestration and retrieval can change the result more than the model.

Scoring04

Score before any edits

Do not fix either answer before scoring it. Check source locations, whether it split absent from ambiguous, and any invented clauses. For real matters, hide the model names and have a lawyer review blind.

Test prompts and what to expect
Four review jobs

Prompts you can run yourself, with the pattern the public evidence suggests. It sums up benchmarks and vendor tests rather than promising a fixed result.

Job
A prompt to try
What the evidence suggests
Likely edge
Playbook review
Review this MSA against the attached playbook. For each issue give the clause, its location, the exact supporting text, the playbook rule, the risk and a proposed revision. Mark not found, ambiguous or conflicting where it applies.
Opus tends to read the agreement as a connected whole and prepare a usable issues list. GPT can follow the format cleanly but is more likely to over-flag or mark a rule met when the clause is absent.
Claude Opus 4.8
Field extraction
Return one JSON object per requirement with a status of met, unmet or uncertain, plus clause, page and an exact evidence quote. Use null when no clause exists. A finding without supporting text is invalid.
GPT-5.5 holds a strict schema well and made narrow, accurate edits in the revision test. Require an evidence span so a clean-looking false positive cannot slip through.
GPT-5.5
Missing-clause check
List every clause the playbook requires that this agreement does not contain. Separate clauses that are absent from clauses that are present but ambiguous. Do not infer missing language.
Opus is the safer default for spotting absent and conflicting language, but both models miss conditions, so keep the review closed to the document and verify.
Claude Opus 4.8
Narrow redline
Change the governing law to Delaware and the notice period to 30 days across this agreement. Make only those edits and flag anywhere the change affects another clause.
GPT-5.5 often makes just the requested edit without rewriting the surrounding text. Ask it to flag knock-on effects, since a literal edit can weaken nearby wording.
GPT-5.5

On the strictest end-to-end test, Harvey's Legal Agent Benchmark, Opus 4.8 passed 10.4% of matters against 2.1% for a GPT-5.5 baseline. The low absolute scores matter more than the ranking: neither finishes most complex matters without at least one material miss.

How to prompt each one
They want different prompts

The best prompt style is not the same for both. Matching the prompt to the model does more for a clean review than the model choice alone.

Claude Opus 4.8 does best when you give it the whole review framework and define uncertainty up front. Anthropic notes that Opus follows literal instructions and can get verbose unless you set the length, so ask for a fixed table and separate fields for what is uncertain and what needs counsel.9

GPT-5.5 does best with a strict output contract that makes an unsupported finding invalid. OpenAI suggests starting with the smallest prompt that keeps the output shape, then tuning the reasoning, verbosity and format against real examples.10 Require an exact supporting quote for every finding, which directly targets its false-positive risk.

A Claude Opus 4.8 prompt: framework and uncertainty

<role>
You are a senior contracts lawyer preparing an issues list.
</role>

<documents>
  <document index="1">
    <source>Agreement</source>
    <document_content>...</document_content>
  </document>
  <document index="2">
    <source>Playbook</source>
    <document_content>...</document_content>
  </document>
</documents>

<instructions>
For each issue give: clause, location, exact supporting text,
playbook requirement, risk and a proposed revision.
Use "not found", "ambiguous" or "conflicting" where they apply.
Do not infer missing language. Keep the table concise.
</instructions>

<query>
Produce the issues table, then list unresolved questions for counsel.
</query>

A GPT-5.5 prompt: a strict output contract

You are reviewing a commercial contract for a US legal team.

Task:
Check each playbook requirement against the agreement.

Output:
Return one JSON object per requirement with:
- status: met, unmet or uncertain
- clause, page, evidence_quote, explanation

Rules:
- Use null when no clause exists
- A finding is invalid without an exact supporting quote
- Do not draft revisions or add legal conclusions
- Treat "not located" and "not present" as different

Weak spots
And how to fix them

Neither model is safe on its own. The useful question is where each one adds risk, and what to change in the prompt or the workflow.

Model
Weak spot
What it looks like
How to fix it
Claude Opus 4.8
Over-analyses
Long answers, or resolving an ambiguity quietly instead of escalating it. Maximum effort can overthink a simple task.
Require a fixed table and word limits, with separate uncertain and counsel-decision fields. Use high effort first and reserve higher settings for hard matters.
GPT-5.5
Clean-looking false positives
A tidy finding that is wrong, a literal edit that weakens nearby wording, or a rule marked met when the clause is absent.
Require an exact quote and location for every finding. Ask for counter-evidence and make it separate not located from not present.
Both models
Invented authority
Made-up citations, missed conditions, and output that looks more reliable than it is.
Keep the review closed to the document where you can. Verify any law against the primary source and add checks for dates and amounts, then a lawyer review.

Which one to choose
Start from your main review job

A quick decision flow. Find the job that matches most of your work, then start with the model on that branch. Every branch ends the same way, with a lawyer checking the result.

Contextual judgment or structured transformation? Issue spotting or a full redline Fixed-field extraction Narrow revision to strict rules Conflicting or missing documents Citation-heavy research Claude Opus 4.8 GPT-5.5 first GPT-5.5 Claude Opus 4.8 Verify every source first Then a lawyer signs off

A starting point, not a rule. Test on your own matters before you commit.

Recommendations
Pick by your review job

If your work is contextual review, issue spotting or a full redline, Claude Opus 4.8 is the better default. It led the public contract-drafting test by a wide margin3, and Harvey saw it review and revise its own drafts before handing them over.6

If your work is fixed-field extraction into a database, start with GPT-5.5, but require an evidence span for every finding and compare it against Opus before you deploy. For a single narrow revision under strict instructions, GPT-5.5 is the safer pick.4

For conflicting documents, missing schedules or ambiguous drafting, choose Opus 4.8 and ask it for an uncertainty register.1 For a very long document set, start with Opus 4.8 and check cost and recall on the real corpus, since GPT-5.5 charges more past 272,000 input tokens.2 For citation-heavy research or client-ready advice, use either only as a first-pass assistant with a lawyer checking every authority.

Bottom line
One split with real limits

If we reduce it to one line: Claude Opus 4.8 is the safer single-model default for reviewing contracts, and GPT-5.5 is the better fit for tightly bounded extraction or instruction-driven edits.

That is a simplification, but it is a fair read of the current evidence. The distinction is not a careful model against an accurate one. Opus can fail to flag ambiguity, and GPT can confidently over-flag or miss a clause that is simply absent.5 Results also move with the prompt, the reasoning setting, parsing, retrieval and how the answers are graded.

Two limits matter most. GPT-5.5 has already been superseded, so treat this as a guide for pinned deployments and same-price choices.13 And no public result supports unsupervised legal use. The American Bar Association's Formal Opinion 512 treats these tools as assistants under the lawyer's own duties of competence and candor, useful for a first pass but never a substitute for review.12 The safest final step is to test on your own matters, with a qualified person checking the output. A fair test needs the same setup for both models: the same document, the same prompt, and the same place to run them, so the result reflects the models and not the tool around them. In practice that is harder than it sounds, since most teams end up running one model in one app and the other in a different one, on two separate subscriptions, which tilts the comparison before the first answer even comes back. The cleaner the setup, the more the difference you see is really Claude Opus 4.8 against GPT-5.5, and not just which one happened to be easier to reach that day.

Test both in one workspace
Right here inside Playgram

That's the practical case for the setup just described, and it's also how the day-to-day work gets easier. When both models sit in one workspace, you can send a document to each, compare the findings side by side, and hand a draft from one model to the other without setting it up again.

Playgram lets you run that same comparison directly: upload the document once, put it in front of both Claude Opus 4.8 and GPT-5.5, and keep the conversation going with either one without re-uploading it or starting over for the second opinion.

The same memory carries across the team too, not just this one comparison, over the latest GPT, Claude and Gemini models in one place15, with the line-up curated so retired models are turned off and new ones added as they ship. It keeps a lawyer in the loop, since it speeds the first pass rather than replacing the review.

Team memory

Shared across everyone and every model.

Project memory

Scoped to a campaign or document set.

Personal memory

Your own working style, kept private.

Fair pricing
Pay per usage, not per seat

Upgrade as needed, and only pay for what you actually use

Save ~17% with the annual plan

Pro

$50/ month

Perfect for small and medium teams

Unlimited users & infinite memory

Multi-LLM chats

Unlimited use of DeepSeek V4 Flash

US & EU data residency

Get started

Enterprise

Get in touch

Unlimited Credits

For organizations with advanced needs

Unlimited users & SSO

Priority Support

Unlimited use of DeepSeek V4 Flash

US & EU data residency

Book a call

30-days money back guarantee

Frequently asked
questions

There is no single winner. Claude Opus 4.8 tends to do better at reading a whole agreement, spotting conflicting or missing clauses and drafting in context. GPT-5.5 tends to do better at tightly defined extraction and narrow edits that must fit a fixed shape. For a mixed workload, many teams use GPT-5.5 to pull fields into a table, then Opus 4.8 to spot issues and draft. Whichever you pick, a lawyer still has to check the result.

No. Every credible public test still shows real error rates, including invented clauses and citations. The American Bar Association treats these tools as assistants under the lawyer's own duties, so they can speed up a first pass but cannot take responsibility for the work. Keep a qualified person in the loop and verify any legal authority against the primary source.

In HAQQ's June 2026 test GPT-5.5 scored highest for accuracy at 8.41 out of 10, with a 3% rate of made-up citations. Even so, it still produced at least one likely fabricated citation, and every model tested had a citation problem. Treat citation-heavy legal research as unsafe from either model unless a person checks every source.

Claude Opus 4.8, on output. Both list $5 per million input tokens, but Opus lists $25 per million output tokens while GPT-5.5 lists $30. GPT-5.5 also charges more once a session goes past 272,000 input tokens, which is easy to reach with a big set of documents. Test cost on your real corpus before you commit.

Pick three to five real matters, such as a standard NDA, an MSA with hidden conflicts, a poor scan and a multi-document amendment chain. Give both the same prompt, the same files and the same settings, and do not fix either answer before scoring. Judge whether each one found the right clauses, flagged what was missing or ambiguous, avoided inventing text, and needed less lawyer editing.

No. Run the same prompt on both and compare the answers, or switch between them mid-conversation. You choose after reading both answers instead of guessing up front.

Related comparisons

GPT-5.5 vs Claude Opus 4.8 for writingPlaygram vs Claude TeamPlaygram vs ChatGPT Business

Send one matter to both
Keep it all in one place

Send the same document to Claude Opus 4.8 and GPT-5.5, keep the context in one place, and see which mix gives a cleaner first pass for your lawyers to check. Set it up in a minute.

Get startedSee the pricing