From the source
Opus 5, the latest Anthropic model, is now available in Harvey.
It builds on prior Anthropic models like Opus 4.8, with broad gains in legal accuracy across litigation and transactional work.
In our early testing, Opus 5 scored 11.7% all-pass on our Legal Agent Benchmark (LAB), a meaningful step up relative to prior Opus models.
Notably, compared to Opus 4.8 at max reasoning, Opus 5 achieved similar performance at lower reasoning levels while generating 26% fewer tokens on average.
This overall efficiency can help legal teams produce outputs that balance quality and succinctness at scale.
In particular, the model excels at transactional and disclosure-heavy work, demonstrating strength in practice areas like corporate governance, corporate M&A, energy, and real estate, where legal teams are often tasked to organize a large body of related material into a clear, well-structured deliverable.
Compared to Opus 4.8, there were also clear gains in practice areas like arbitration and international disputes, where careful analysis of detailed factual records is critical.
…
