GPT-5.4, Now Live in :Harvey:
Announcing GPT-5.4 in Harvey.
Today, we are making GPT-5.4 available in Harvey. GPT-5.4 is OpenAI's latest frontier model, building on the strengths of GPT-5.2 in structured reasoning and document-grounded analysis while delivering meaningful improvements in substantive accuracy, drafting format, and organizational clarity.
Early access evaluations point to gains in transactional accuracy, drafting precision, and the ability to handle multi-step legal tasks that require careful attention to contractual detail. On our BigLaw Bench evaluation suite, GPT-5.4 scored 91.0% — a step up from GPT-5.2’s score of 89.8%. The model received perfect scores for BLB’s risk assessment and compliance tasks; with deal management, drafting, and analysis of litigation filings as additional standout areas. We expect GPT-5.4 to excel in drafting-intensive work where getting the details right is essential.
During early access testing, our Applied Legal Research team highlighted several strengths. The model is impressively fast and comprehensive, with strong underlying substance and straightforward readability. Notably, GPT-5.4 gets straight to the task at hand — it avoids preamble writing and throat-clearing behavior that we’ve observed in prior models, instead diving directly into the legal question posed. On organizational tasks, evaluators highlighted GPT-5.4's intuitive structure and appropriate level of detail as clear strengths.
“GPT-5.4 continues a clear trajectory of improvement for legal reasoning within the GPT family. The model's gains in drafting format and organizational structure are the kind of practical progress that matters for day-to-day legal work. This launch reflects targeted improvements over GPT-5.2 for legal practitioners.”
Niko Grupen
Head of Applied Research at Harvey
We'll be rolling out GPT-5.4 in the model selector to eligible clients in the US and EU in the coming days. AU availability will be announced as in-region capacity becomes available.
