New Global Legal AI Benchmark Launched

Harvey has introduced BigLaw Bench: Global, a major expansion of its BigLaw Bench benchmarking framework designed to evaluate AI models' performance on legal tasks across multiple jurisdictions. This release represents a significant step toward ensuring that large language models can reliably execute legal workflows in localized legal contexts.

What's Changing

The new global dataset more than doubles the size of Harvey's public-facing benchmark, adding comprehensive evaluations for three jurisdictions:

• United Kingdom
• Australia
• Spain

Built in collaboration with Mercor, a leading expert data company, BLB: Global moves beyond generic AI evaluation to test models on jurisdiction-specific legal nuances, terminology, and regulatory requirements.

Key Focus Areas

Harvey identified six major task categories that legal teams rely on AI to perform:

• Drafting: Production of legal documents, from analysis memos to contracts
• (Additional categories enumerated in original announcement)

The benchmark addresses a critical gap: while leading foundation models now consistently achieve around 90% performance on core legal tasks in general benchmarks, their accuracy can degrade significantly when applied to localized legal contexts. BLB: Global is designed to identify and remediate these localization failures.

Why It Matters

As legal AI adoption expands globally, models must accurately reflect local legal principles, regulatory frameworks, and professional norms. This benchmark enables Harvey's customers—and the broader legal AI ecosystem—to verify that AI systems can execute workflows with accuracy and consistency across different jurisdictions. The framework will help inform model selection and deployment decisions for law firms operating internationally.