Claude Sonnet 5, Anthropic's latest model, is now live in Harvey, delivering improved legal task performance for enterprise use.
Anthropic's Claude Sonnet 5, the latest iteration of its mid-tier large language model, is now available in Harvey, the AI-driven legal assistant platform. Officially launched on June 30, 2026, Sonnet 5 aims to deliver significant improvements in legal accuracy and multi-step task execution, targeting both litigation and transactional workflows.
According to Harvey, Sonnet 5 achieved a 5.8% all-pass score on the platform's Legal Agent Benchmark (LAB), a stringent evaluation designed to simulate complex legal tasks at law firms.该模型在起草任务方面表现出色，特别是在能源和自然资源、房地产和资本市场方面。 On the BigLaw Bench, a separate benchmark focused on large law firm use cases, Sonnet 5 posted an impressive 91.3% score—surpassing all previous Sonnet and Opus models.显着的优势包括风险评估、合规性和案例管理。
“与 Sonnet 4.6 相比，Sonnet 5 在法律质量方面取得了有意义的飞跃，”Harvey 应用研究主管 Niko Grupen 评论道。 “在早期测试中，它比其前身更准确、更精确，用更少的文字提供了更有力的答案。” However, Grupen also noted that the model still faces challenges with dense, specialized analytical tasks, such as tax and structured finance, and remains imperfect in completing every step of complex multi-step tasks without error.
Claude Sonnet 5 represents Anthropic's effort to make advanced 'agentic' AI capabilities—autonomous task execution, tool usage, and enterprise workflow integration—more accessible and affordable. Positioned between the costlier Mythos- and Fable-class models and the older Sonnet 4.5, it balances performance with practicality for mainstream enterprise adoption.该模型还遵循 Anthropic 的网络验证计划，确保增强现实世界部署的可靠性和安全性。
Harvey 计划在未来几天内向符合条件的美国客户推出 Sonnet 5，并很快在欧盟和澳大利亚推出。 As Anthropic continues expanding its AI offerings across platforms like AWS, Microsoft Foundry, and soon Google Vertex AI, Sonnet 5 could play a pivotal role in bridging the gap between frontier AI capabilities and everyday enterprise AI deployment.