Anthropic ships Claude Opus 4.8 with effort controls and dynamic workflows

Anthropic released Claude Opus 4.8, an upgrade over 4.7 that holds the same price while posting gains across coding, agentic, and reasoning benchmarks. Early partners highlight sharper judgment in Claude Code, stronger tool-calling efficiency, and a top score of 84% on the Online-Mind2Web browser-agent benchmark. Anthropic claims the model is roughly four times less likely than its predecessor to let flaws in its own code slip through unflagged, and its alignment team reports lower rates of deception and misuse cooperation than 4.7.

The launch bundles several product changes. A new effort control on claude.ai and Cowork lets users trade speed for depth, defaulting to high effort with optional extra and max tiers. Claude Code gains dynamic workflows in research preview, letting Claude plan tasks and run hundreds of parallel subagents for jobs like codebase-wide migrations gated by the existing test suite. The Messages API now accepts system entries inside the messages array, so developers can update instructions mid-run without busting the prompt cache, and fast mode for Opus 4.8 is three times cheaper than on prior models.