Harvey integrates GPT-5.4, OpenAI's latest model; scores 91% on legal benchmarks
GPT-5.4 Now Available in Harvey
OpenAI's latest frontier model, GPT-5.4, is now live in Harvey, an AI platform purpose-built for legal teams. The integration brings meaningful improvements in substantive accuracy, drafting precision, and organizational clarity compared to the prior GPT-5.2 model.
Performance Improvements
Early access evaluations demonstrate tangible gains across legal workflows:
- Overall accuracy: 91.0% on Harvey's BigLaw Bench evaluation suite, up from GPT-5.2's 89.8%
- Perfect scores: Risk assessment and compliance tasks
- Standout areas: Deal management, drafting work, and litigation filing analysis
- Notable behavior: The model avoids preamble and filler text, diving directly into legal questions
Key Strengths for Legal Work
Harvey's Applied Legal Research team highlighted several practical advantages:
- Drafting-intensive work: Excels where attention to contractual detail is critical
- Speed and comprehensiveness: Fast analysis with strong underlying substance
- Organizational structure: Intuitive layout with appropriate levels of detail for legal documents
- Readability: Straightforward output that practitioners can quickly act on
Rollout Timeline
GPT-5.4 will be available in Harvey's model selector for eligible clients in phases:
- US and EU: Rolling out in the coming days
- Australia: Availability will be announced as regional capacity becomes available
Legal teams using Harvey can expect to leverage these improvements immediately for transactional work, contract analysis, and complex multi-step legal tasks requiring careful attention to detail.