Anthropic's Claude Opus 4.8 delivers 2.5x speed and smarter agents

Anthropic today announced Claude Opus 4.8, an upgraded version of its flagship model. The new release brings significant improvements across benchmarks, particularly in agentic tasks, coding, and reasoning. Users on claude.ai now have granular control over the effort Claude puts into tasks, while Claude Code introduces dynamic workflows for tackling large-scale problems. Fast mode for Opus 4.8 operates at 2.5× the speed and is three times cheaper than it was for previous models, making high-performance AI more accessible.

Early adopters highlighted superior reliability and judgment. The model achieved the highest score on Anthropic's Legal Agent Benchmark, becoming the first to break 10% overall on the all-pass standard. On CursorBench, it exceeded prior Opus models at every effort level. For computer-use tasks, Opus 4.8 scored 84% on Online-Mind2Web, a meaningful jump over both Opus 4.7 and GPT-5.5. Testers from Devin, Databricks' Genie, and Hebbia reported faster, more consistent outputs and proactive issue detection. The model also fixed verbosity and tool-calling issues seen in Opus 4.7.