Anthropic released Claude Opus 4.8 on May 28, 2026. While the version number suggests an incremental update, this release fundamentally improves model reliability—the critical metrics determining whether AI agents can run unattended for extended periods. Key improvements include: a 4x reduction in code defects the model fails to notice; fixes for silently skipped tool calls that corrupt long agent trajectories; better history compression recovery preventing long-horizon runs from derailing; dynamic workflows allowing models to coordinate hundreds of parallel subagents for codebase-scale work; adaptive thinking deciding per-turn whether reasoning is needed; and a fast mode running 2.5x faster at 3x cheaper than 4.7. Regular mode pricing unchanged. Unlike benchmark improvements, reliability gains directly impact whether autonomous agent systems work in production. The accelerated release cadence (6 weeks from 4.7) signals Anthropic's strategic shift from capability-focused to reliability-focused releases.