GPT-5 is Now Live in Spellbook

Marc Manara, Head of Startups at OpenAI noted: “Working with Spellbook on GPT-5 is the latest example of how their feedback and early testing will help us identify the API's optimal impact where it matters most for transactional lawyers.”

Nuanced Issue Spotting

The number one thing we’re excited by is the level of nuance in GPT-5’s issue spotting. It has been able to cross-reference information and spot issues in ways that Claude 4 and GPT-4o haven’t. For example, we’ve never seen Sonnet or GPT-4o flag the issue below: the interest rate was stated in a way that wasn’t compliant with Ontario’s Interest Act.

GPT-5 was able to cross-reference the contract jurisdiction, the interest rate terms and the Interest Act.

We also found that GPT-5 was much more fluent in security and compliance language—again able to cross-reference standards in different jurisdictions. Below we see it expand a DPA to an appropriate level of detail and transparency in an enterprise agreement.

‍

Surgical Legal Document Revision

We’re also excited by GPT-5’s performance in surgical Word document revision tasks where less legal judgement is required. For instance—using a termsheet to modify an agreement. A lot of Gen AI experiences spit out entirely new documents, but transactional lawyers are rarely drafting from scratch. They typically have a precedent that they want to modify. And that precedent is not plain text—it’s in Word with tables and other rich formatting.

Contracts like Share Purchase Agreements can be 100+ pages long. Making significant revisions to them requires a lot of jumping around, consistency checking, and making sure numbers add up.

It’s a deep challenge to get these tasks performing well with previous models.

Here are a few examples of where we've seen GPT-5 outperforming:

Surgical Paragraph Editing

Claude 4 (left) struggles to make surgical edits in this Share Purchase Agreement, striking an entire paragraph and re-adding it. This makes it difficult to understand what changed. GPT-5 (right) is better able to make surgical edits while keeping untouched text intact.

Field Filling Completeness

A more subtle case we found: Claude 4 (left) would sometimes leave in optional term brackets. It left “[2.8(f) and/or 2.8(k)]” untouched even though it was instructed to prepare a final document. GPT-5 (right) successfully struck the square brackets, including the optional language.

Understanding Table Structures

In the example below, we compared GPT-5 with Claude 4 in Spellbook Associate to update a Master Services Agreement based on a termsheet:

With GPT-5, we are seeing dramatic improvements for revision tasks across the board. One of our top predictions is that there will be a lot more workflows based on surgical document revision launched this year—rather than just outputting isolated text.

More to Come

This is just the beginning of our journey with GPT-5. We’ll be unlocking new use cases and posting further evaluations as we get more feedback from our customers.

GPT-5 is live now in Spellbook Associate for contract revision, and is being gradually rolled out in our contract review functionality.