SDSignal Desk

Harvey turns legal context into stronger drafts with GPT-6 Astra

Sep 23, 2026, 5:00 AM · OpenAI

Image: OpenAI

Harvey says GPT-6 Astra lets it feed more matter context into drafting and return more structured legal documents — so lawyers spend less time on formatting and more on strategy.

Why it matters

Legal AI lives or dies on whether drafts reflect the actual matter — court filings, firm docs, case law — and come out structured enough to edit, not rewrite.

OpenAI’s customer story positions Astra as the model that finally lets Harvey push more of that context through and get cleaner formatting back. For firms already deploying Harvey on litigation and M&A workflows, that’s a quality jump inside an existing secure stack.

From the desk

We’re covering this as a vertical-tools story: context windows and structure, not courtroom autonomy.

Harvey helps law firms and in-house teams deploy AI across complex legal workflows. With GPT-6 Astra, the company reports substantial improvements in document formatting and context awareness versus other models it tried, producing more complete documents that better reflect the underlying material. Cofounder and President Gabe Pereyra’s line is straightforward — more context in, better structured outputs out.

The memory panel is the workflow detail that matters. Lawyers can encode preferences — numbered lists, prioritizing EDGAR, color-coding issues by priority — and see those preferences beside source material and the draft memo. Astra’s ability to process more context is what makes that combination usable instead of truncated.

Useful AI in law looks like this when it works: amplify judgment, don’t replace it. Structured drafts that respect firm preferences free senior time for strategy and risk calls. The downside is familiar and serious. Stronger formatting can mask weak reasoning; confidence in a polished memo is not the same as correctness on the law. Privilege, confidentiality, and hallucination risk don’t vanish because the headings look right. Firms still need human review gates proportional to the stakes of the filing.

I’m watching whether Astra’s context gains show up as fewer rewrite cycles in real matters, and whether preference memory becomes a competitive moat or table stakes across legal AI vendors.

Context

OpenAI Startup customer story, Sep 23, 2026, on Harvey’s use of GPT-6 Astra for context-aware legal drafting via the API. North America technology startup.

Who feels it

Law firms / in-house counsel
Higher-context drafting may cut formatting churn; keep mandatory attorney review on anything client- or court-facing.
Legal AI vendors
Preference memory plus long-context models is becoming the product shape customers expect.
Compliance / risk teams
Better-looking drafts raise the need for evals on legal accuracy, citation fidelity, and privilege boundaries — not just structure.
OpenAI platform watchers
Another Astra vertical proof point aimed at professional services workloads that live on documents.

What to watch

  1. Firm-reported reductions in draft revision cycles after Astra rollout.
  2. Independent evals of legal factuality and citation quality versus formatting gains.
  3. How preference/memory features handle multi-lawyer teams with conflicting style rules.
  4. Competitive responses from other legal AI platforms on long-context drafting.

Read the original

Continue at the source.

OpenAI

Companies: OpenAI