Claude Sonnet 4.6 Now Available in Harvey

Harvey has made Claude Sonnet 4.6—Anthropic's latest model in the Sonnet family—available to eligible clients within its legal AI platform. Sonnet 4.6 combines strong reasoning capabilities with a faster, more efficient architecture designed for practical problem-solving and instruction-following.

Performance Metrics

On Harvey's proprietary BigLaw Bench evaluation suite, Sonnet 4.6 achieved an 87.6% overall score, with 35% of tasks receiving perfect scores. Performance was balanced across practice areas, with equivalent average scores for both litigation and transactional legal work.

Key Strengths

Harvey's Applied Legal Research team identified several areas where Sonnet 4.6 excels:

• Numerical precision: Delivers more accurate specific figures (e.g., dollar amounts) when explicitly instructed compared to prior Claude models
• Ideation tasks: Shows clear advantages in thoroughness for trial exhibit preparation and open-ended litigation strategy questions
• Code analysis: Excels at root cause analysis in Claude Code, correctly tracing bugs through code history and integrating parallel tool calls to solve problems that even Opus 4.6 missed

Important Usage Pattern

Evaluators noted an important nuance: Sonnet 4.6 performs best when given explicit direction (retrieve exact figures, follow a defined rubric, produce structured comparisons). When left to determine scope independently, it tends to remain at higher summary levels, sometimes missing fine-grained details—a gap most visible on multi-document synthesis tasks.

Rollout

Harvey is rolling out Sonnet 4.6 in the model selector to eligible clients, with broader availability expected to follow.