Claude Sonnet 5, Now Live in :Harvey:
Announcing Sonnet 5 in Harvey.
Today, we are making Claude Sonnet 5 available in Harvey. Sonnet 5 is Anthropic's latest Sonnet model, building on Sonnet 4.6 with broad gains in legal accuracy and output quality across transactional and litigation work.
Sonnet 5 scored 5.8% all-pass on Harvey’s Legal Agent Benchmark (LAB), a strict evaluation that mirrors how multi-step legal work is executed at a law firm, from assignment through review. The model showed strengths across a number of practice areas, including energy and natural resources, real estate, and capital markets — with drafting as its strongest LAB task type.
On BigLaw Bench, Sonnet 5 scored 91.3% — the highest we've recorded across both Sonnet and Opus models. Its consistent performance across litigation and transactional work, combined with its cost efficiency, makes it especially well-suited for high-throughput tasks. Risk assessment and compliance, case management, and transactional drafting were standout BigLaw Bench task types.
“Sonnet 5 brings a meaningful jump in legal quality over Sonnet 4.6. In early testing, it was both more accurate and more precise than its predecessor, delivering stronger answers in fewer words.”
Niko Grupen
Head of Applied Research, Harvey
While drafting and transactional tasks are strong, dense, specialized analytical tasks such as tax and structured finance remain more challenging, as they do across frontier models. There's also room to improve on the strictest agentic measure — completing every step of a long, multi-step task without a single miss.
We’ll be rolling out Sonnet 5 in the model selector to eligible customers in the US in the coming days. EU and AU availability will be announced shortly.
