On May 28, 2026, Anthropic released Claude Opus 4.8, an upgrade to Opus 4.7 positioned as a stronger collaborator on coding, agentic tasks, and professional work. The announcement says the model is around four times less likely than its predecessor to let flaws in code go unremarked, scores 84 percent on the Online-Mind2Web browser and computer-use benchmark (a jump over both Opus 4.7 and GPT-5.5), and posts the highest score Anthropic has recorded on its Legal Agent Benchmark - the first model to break 10 percent on the all-pass standard.
Beyond raw scores, the release leaned on workflow features. A new “dynamic workflows” capability in Claude Code lets the model spin up parallel subagents to tackle large-scale problems, and an effort control on claude.ai lets users trade quality against token usage. Anthropic also says fast mode runs at 2.5 times normal speed and costs three times less than fast mode on previous models. Standard pricing was unchanged at $5 per million input tokens and $25 per million output tokens.
For business readers, Opus 4.8 is a reminder that frontier progress in 2026 is increasingly measured in agentic reliability - whether a model asks the right questions, catches its own mistakes, and pushes back on bad plans - rather than in single benchmark headlines. Those are the traits that determine whether an AI can be trusted with multi-step work.