Anthropic Unveils Claude Opus 4.8: Enhanced Honesty, Dynamic Workflows, and Competitive Benchmarks

Anthropic officially released Claude Opus 4.8 on May 28th, positioning it as their most powerful public-facing large language model, succeeding Opus 4.7. The new iteration boasts general improvements across all benchmarks and maintains its pricing structure at $5 per million input tokens and $25 per million output tokens for standard usage. Opus 4.8 introduces several new functionalities, including enhanced control over model “effort” and a new “Fast Mode” offering 2.5x speed at double the standard cost, notably being three times cheaper than previous fast mode iterations. A significant addition for enterprise users is Dynamic Workflows within Claude Code, available for Enterprise, Team, and Max plans, designed to tackle complex, end-to-end tasks by dynamically generating scripts to execute hundreds of sub-agents in parallel.

Performance-wise, Opus 4.8 surpasses its predecessor, Opus 4.7, across all evaluated benchmarks. Competitive analysis shows GPT 5.5 holding a slight 4% lead in Agentic Terminal Coding, while Google’s Gemini 3.1 Pro is noted to be significantly lagging, particularly in agentic tasks. A key focus for Opus 4.8 is improved honesty, with Anthropic claiming the model is four times more likely to admit when it cannot solve a problem or lacks information, significantly reducing “misaligned behavior.” However, the Dynamic Workflows feature stirred controversy post-launch due to its initial activation mechanism: simply typing the word “workflow” in Claude Code would trigger the expensive multi-agent mode, leading to unintended high token consumption. Anthropic has since addressed this, allowing deactivation via prompt or configuration. While some early user feedback suggests no “substantial improvement” over Opus 4.7 and even perceived regressions in specific development tasks, the model’s overall capabilities and maintained pricing are generally well-received.