Anthropic released Claude Opus 4.8, the latest iteration of its flagship model, which builds on Opus 4.7 with measurable benchmark improvements. The upgrade arrives without price changes and introduces enhanced collaboration capabilities. Early testing shows the model demonstrates better judgment in agentic tasks, asking more relevant questions and catching mistakes that previous versions missed.

Claude.ai users gain effort control, selecting how much computational intensity the model applies to responses. Meanwhile, Claude Code adds dynamic workflows that enable parallel subagent execution for large-scale tasks. Fast mode now runs at 2.5× speed while costing three times less than prior models. These features target developers building autonomous systems who need reliable, long-running agent performance.

Benchmark results highlight significant gains across coding, reasoning, and professional workflows. On the Super-Agent benchmark, Opus 4.8 is the only model completing every test case end-to-end. The model achieves 84% on Online-Mind2Web for computer-use tasks and scores highest on Legal Agent Benchmark testing. Improvements include better citation precision and reduced unsupported claims—early testers report it's about four times less likely to let flawed code pass unremarked.

For enterprise users, the model delivers substantial cost savings: 61% cheaper token costs for PDF and diagram reasoning in Databricks' Genie platform. Alignment assessments show lower rates of deceptive behavior compared to Opus 4.7. The combination of performance, honesty, and cost efficiency positions this release as a meaningful step forward for production AI agents.