Anthropic launches Claude Opus 4.8 with Effort Control

Anthropic released Claude Opus 4.8 on May 28, 2026. The update improves coding accuracy, agentic workflows and practical reasoning, and introduces a user-facing Effort Control setting and a Dynamic Workflows research preview. The model is available at the same standard price across claude.ai, the Claude API and major cloud platforms.

Anthropic and early testers reported measurable gains over Opus 4.7. Internal performance tests cited by the company indicate the new model is about four times less likely to miss flaws in code it generates and is less prone to making unsupported claims. Alignment testing showed higher scores on prosocial measures and lower rates of misaligned behavior than the previous release. Testers noted increased reliability, sharper judgment and clearer statements about task progress.

Effort Control lets users choose how much cognitive effort the model applies. Options include Low, which runs faster and uses fewer rate-limited tokens, and Max for the most thorough responses. Anthropic sets the default to High to balance response speed and quality.

Dynamic Workflows is available as a research preview in Claude Code. The feature enables Claude to plan large projects, run and coordinate hundreds of parallel subagents, and verify their outputs. Anthropic says the workflow can handle codebase-scale migrations spanning hundreds of thousands of lines of code.

Developers received a Messages API update that allows system instructions to be inserted mid-conversation without breaking prompt cache, a change intended to support long-lived, stateful interactions. Anthropic also introduced a faster execution tier for Opus 4.8 called Fast Mode, which runs at roughly 2.5× speed and carries its own pricing.

Standard pricing for Opus 4.8 remains $5 per million input tokens and $25 per million output tokens. Fast Mode is priced at $10 per million input tokens and $50 per million output tokens; Anthropic said that rate is about three times cheaper than prior fast-mode pricing. The model is available immediately as claude-opus-4-8 via the Claude API and through major cloud providers in addition to claude.ai.

Anthropic said it plans lower-cost variants with similar capabilities and is preparing Mythos-class models for wider release after finishing additional cybersecurity work under Project Glasswing. Anthropic posted on X: “Available today at the same price.” Enterprises and developers can begin testing Opus 4.8 on complex coding and agentic workflows now that the update is live.