Harvey transforms legal work with Claude

Harvey helps law firms and Fortune 500 enterprises revolutionize their approach to contract analysis, due diligence, and litigation by integrating Claude's advanced reasoning into their domain-specific AI platform.

Key results with Claude:

• Deployed Claude across the platform in under one month
• Achieved one of the highest performance scores on Harvey's proprietary BigLaw Bench evaluation
• Delivered statistically significant improvements on complex workflows requiring long-context reasoning
• Deployed Claude at enterprise scale and with enterprise-grade security and privacy.

Legal AI demands exceptional accuracy and deep reasoning

Building AI for legal professionals requires processing thousands of documents with deep reasoning while maintaining the precision that legal work demands. With customers ranging from global law firms to Fortune 500 enterprises, Harvey needs AI models that excel at contract analysis, due diligence, compliance, and litigation support.

"Our secret sauce is this environment where we have domain experts embedded into teams and processes," said Niko Grupen, Head of Applied AI at Harvey. "We have legal researchers who are former Big Law attorneys sitting alongside AI researchers and engineers. They whiteboard together from concept, mapping legal problems to model problems."

This deep domain expertise drives Harvey's rigorous approach to model selection. "Evaluation of models for domain-specific tasks is one of our core strengths at Harvey," said Grupen. The company tests every model through BigLaw Bench, their proprietary benchmark that measures AI's ability to complete real-world legal tasks. These outputs are graded for both substance—including accuracy, materiality, and hallucination rates—and form, such as tone, length, and formatting.

Harvey chose Claude for superior long-context performance

When Harvey evaluated models for their platform expansion, Claude delivered exceptional performance. "We found that Claude is one of the highest performing models on our evaluations," said Grupen.

What distinguished Claude was its ability to handle long-context reasoning—critical for legal work involving extensive documents. "We've seen it perform really well on complex workflows that require long-context reasoning," Grupen explained. "The ability to effectively reason over a maximally-utilized context window remains a significant challenge, and Claude performs really well at upper token thresholds."

Harvey's three-pillar evaluation process tested models through BigLaw Bench, real-world product environments, and unstructured assessments by both AI and legal researchers. Claude delivered statistically significant improvements on complex workflows where long-context reasoning proves essential. This performance, combined with Harvey's commitment to offering their customers choice, made Claude an ideal addition to their platform.

Claude available across Harvey’s platform

Harvey customers can select Claude models across key product surfaces — including Assistant, Vault, and Workflows — to help review and draft documents faster, more efficiently, and in the voice and tone of their organization or end customers. The Workflows product, in particular, is well-suited for Claude's instruction-following capabilities and ripe for customization.

"We're taking complex legal tasks and turning them into end-to-end, interactive journeys. Instead of wrestling with natural-language prompts, you can engage with a Harvey agent that guides the process — proactively asking for input, asking the right follow-up questions, handling the heavy lifting behind the scenes, and surfacing intermediate work product for review.”

This human-in-the-loop approach is pivotal: "At every checkpoint you can step in—add, edit, delete, or reshape content—before giving it your stamp of approval." Grupen explains, "The Harvey system then comes up with the final work product from there."

Rapid deployment meets enterprise security demands

Harvey integrated Claude into their platform in under a month—remarkable speed for an enterprise deployment. "Initial feedback from beta testers has been quite strong," Grupen noted.

The successful deployment required meeting stringent enterprise requirements: security and privacy standards, strict compliance protocols, and rolling out significant capacity across multiple regions (in accordance with local data processing requirements). This swift implementation reflected both Harvey's technical capabilities and Anthropic's support throughout the process. "Working with the Anthropic team has been a real delight. We've had support from day one—from leadership all the way to the implementation team," Grupen shared. "We're a company that moves at the speed of light, and having a partner that's in the trenches with us goes a long way."

With the addition of Claude to Harvey’s platform, legal teams can tackle their most complex matters with enhanced capabilities. Whether conducting due diligence on massive data rooms, drafting sophisticated agreements, or navigating intricate regulatory requirements, lawyers across Harvey's global customer base have AI that understands the nuances their work demands.

Building the future of AI-powered legal services

Looking ahead, Harvey plans to expand Claude's role in areas where deep reasoning matters most. They’re particularly excited about the growing interplay between models and agents, especially as they’re manifested in products like Workflows. "This ecosystem of models and their instruction-following capabilities become really important as we innovate," said Grupen. "We're optimistic about how we can evolve and grow together."