James Ding
Could 28, 2026 18:04
Claude Opus 4.8 achieves document 10.4% on LAB, 91.1% on BigLaw Bench, redefining AI efficiency in authorized workflows.
Anthropic has launched Claude Opus 4.8, its newest flagship giant language mannequin, on Could 28, 2026. Designed for advanced authorized workflows, the mannequin has set new efficiency benchmarks, attaining a record-breaking 10.4% rating on Harvey’s Authorized Agent Benchmark (LAB) and 91.1% on the BigLaw Bench.
These metrics characterize vital enhancements over its predecessor, Opus 4.7, which scored 7.1% on LAB. The LAB check evaluates end-to-end completion of intricate authorized duties below a stringent all-pass system. In the meantime, the BigLaw Bench outcomes present 43% excellent scores and 88% of outcomes scoring above 0.80, with standout efficiency in deal administration, compliance, and threat evaluation.
In keeping with Niko Grupen, Head of Utilized Analysis at Harvey, Opus 4.8’s distinctive means to self-review and refine its outputs has contributed to its superior drafting accuracy. “Opus 4.8 is the primary mannequin to interrupt 10% on LAB’s robust all-pass scoring. Its review-and-revise conduct results in stronger efficiency on drafting duties,” he acknowledged.
The mannequin additionally excels in authorized accuracy throughout observe areas, together with figuring out appropriate case captions, events, and statutory provisions whereas making use of doctrinal frameworks precisely. Nevertheless, evaluators famous some room for enchancment in dealing with ambiguity and flagging areas requiring human assessment.
Past authorized duties, Opus 4.8 introduces enhanced coding capabilities via its “dynamic workflows” instrument, which helps large-scale codebase migrations. It additionally options quicker processing speeds—2.5× quicker in quick mode in comparison with earlier iterations—at a lowered value. These updates solidify Opus 4.8’s place not simply in authorized tech however as a flexible AI platform for broader enterprise functions.
Opus 4.8 will probably be out there to eligible purchasers within the U.S. and the EU within the coming days, whereas availability in Australia is predicted to observe. The mannequin is accessible through claude.ai, Anthropic’s API platform, and main cloud suppliers.
Anthropic’s launch of Opus 4.8 comes as the corporate prepares to develop entry to its extra superior “Claude Mythos” collection, aimed toward cybersecurity and high-safeguard use instances. With Opus 4.8’s launch establishing new efficiency benchmarks, it alerts Anthropic’s intent to stay a pacesetter within the aggressive AI house.
Picture supply: Shutterstock
