Opus 4.8, Now Live in :Harvey:
Announcing Claude Opus 4.8 in Harvey.
Today, we are making Claude Opus 4.8 available in Harvey. Opus 4.8 is Anthropic’s latest frontier model, building on the strengths of Opus 4.7 with continued gains in substantive legal accuracy and practical output quality across complex workflows.
On Harvey's Legal Agent Benchmark (LAB), which measures end-to-end completion of complex legal tasks under a strict all-pass standard, Opus 4.8 scored 10.4% — up from Opus 4.7's 7.1% — and is the first model to break 10% on LAB’s strict all-pass scoring system.
On BigLaw Bench, Opus 4.8 scored 91.1%, also a new high for the Claude family. The model achieved 43% perfect scores on BigLaw Bench, with 88% scoring at or above 0.80. Deal management and risk assessment and compliance were particular standouts, with the model reaching perfect or near-perfect marks across both categories.
Our lawyer evaluators highlighted the model’s legal accuracy across practice areas — identifying correct case captions, parties, statutory provisions, and core legal issues, and applying the right doctrinal frameworks. The model also demonstrated strong calibration on tone and output length, delivering responses appropriately sized for the task at hand.
“Opus 4.8 is the first model to break 10% on LAB’s tough all-pass scoring. Qualitatively, we've noticed Opus 4.8 often interrogates its own outputs and edits them before returning an answer. That review-and-revise behavior leads to stronger performance on drafting tasks.”
Niko Grupen
Head of Applied Research at Harvey
While core legal analysis is strong, early access evaluators noted room for improvement in how the model handles ambiguity, particularly around surfacing judgment calls to the user and flagging areas that need human review.
We'll be rolling out Opus 4.8 in the model selector to eligible clients in the US and EU in the coming days. AU availability will be announced shortly.
