Harvey Adds Claude Opus 4.6 to Legal AI Platform

Harvey has made Claude Opus 4.6 available on its platform, delivering improved accuracy and reasoning for legal work.

The February 5, 2026, release extends Opus 4.5's agentic problem-solving strengths while delivering improvements across complex legal workflows. Harvey clients in the US and EU will gain access to the model through the model selector in the coming days, with Australian availability to follow.

Legal professionals require AI systems that handle intricate, multi-step tasks with precision. Opus 4.6 addresses the legal sector's need for substantive accuracy, thorough source grounding, and nuanced handling of transactional matters where details determine outcomes. The model represents a step toward AI systems that can work alongside lawyers on their most challenging matters.

Performance Benchmarks

On Harvey's BigLaw Bench evaluation suite, Opus 4.6 achieved a 90.2% score, the highest yet for Claude models. This represents a significant milestone in legal AI performance evaluation. The model received perfect scores on 40% of tasks, demonstrating strong legal reasoning across litigation and transactional practice areas.

The evaluation suite tests AI models on tasks that mirror real-world legal work. Deal management, risk assessment, and corporate strategy tasks stood out as particular strengths. These areas require models to understand complex commercial relationships, identify potential legal risks, and navigate multi-party transactions.

Early access evaluations showed gains in substantive accuracy, source grounding, and the ability to handle multi-step legal tasks with nuance and precision. The model's performance suggests it can assist lawyers with work that traditionally required extensive manual review and analysis.

Research Capabilities

Harvey's Applied Legal Research team identified several key improvements during early access testing. The model produces outputs with high substantive accuracy and improved analytical depth on complex, multifaceted legal questions. This matters for legal professionals who need AI systems to handle nuanced inquiries that span multiple legal doctrines or jurisdictions.

Opus 4.6 has resolved stylistic issues from earlier versions. Final answers are cleaner, without the preamble text that sometimes appeared in prior Claude models. This change makes outputs more efficient for lawyers who need direct answers without unnecessary context.

The model's source grounding is notably thorough. In one complex research query, the model generated over 120 inline citations, each tied to specific passages in the source material. This level of citation density helps legal professionals verify the AI's reasoning and trace conclusions back to authoritative sources.

For research-intensive work, due diligence projects, and complex transactional matters, the model's citation capabilities reduce the time lawyers spend verifying AI-generated analysis. Legal professionals can quickly check whether the model's conclusions align with the source documents it references.

"Opus 4.6 reflects remarkable progress in agentic capabilities," said Niko Grupen, Head of Applied Legal Research at Harvey. "We've seen meaningful gains in multi-step reasoning and task completion. This release is another step toward AI that can work alongside lawyers on their most challenging matters, and Harvey clients are once again among the first to benefit."

Response Calibration

The model demonstrates an interesting pattern in how it calibrates responses under uncertainty. When Opus 4.6 is confident, it delivers precise, high-quality outputs of appropriate length. This produces efficient results for straightforward legal questions where the model has strong signal from its training.

When less certain, it shows a tendency to over-contextualize responses, producing unnecessary tokens along the way. This behavior suggests the model errs on the side of providing more context when it lacks confidence in a direct answer. While this may add verbosity, it can help legal professionals understand the model's reasoning process.

Harvey continues to refine how it deploys extended thinking for legal workflows that need clear, succinct output. The platform expects Opus 4.6 to excel in research-intensive work, due diligence, and complex transactional matters where getting the details right is essential. These use cases benefit from the model's improved accuracy and source grounding capabilities.

Rollout Timeline

Eligible clients in the United States and European Union will receive access to Opus 4.6 through Harvey's model selector in the coming days. The company will announce Australian availability shortly. This geographic rollout prioritizes Harvey's largest markets first.

The phased rollout allows Harvey to monitor performance and gather feedback from legal professionals working on diverse practice areas and matter types. Law firms and corporate legal departments will test the model across various workflows before broader deployment.

About Harvey

Harvey is an AI platform built specifically for legal professionals. The company provides AI-powered tools for legal research, document analysis, and workflow automation. Harvey works with law firms and legal departments to deploy Claude models for complex legal tasks requiring accuracy and precision.

About Anthropic

Anthropic develops AI systems, including the Claude family of models. The company released Claude Opus 4.6 as its latest frontier model, focusing on agentic problem-solving and complex reasoning tasks. Anthropic provides Claude through API access, developer platforms, and partnerships with specialized platforms like Harvey for legal workflows.

Read more about Harvey here:

Harvey AI

company