Descriptions:
Nate Herk delivers a day-one breakdown of Claude Opus 4.8 from Anthropic, covering what changed from Opus 4.7 and how Claude Code users should adapt their workflows. Released May 28, 2026, Opus 4.8 is priced identically to 4.7 on input and output tokens, but brings notable improvements: sharper honesty about task progress (the model is less likely to falsely claim completion), more sustained autonomy on long-running tasks, warmer collaborative tone, and better tool calling and token efficiency.
A central focus is the new effort-level system — a slider ranging from low to ultra (extra-high plus dynamic workflows) that controls how much compute and reasoning the model applies. Herk argues that effort level is now the single most important lever for Claude Code users, and that many common frustrations with Opus 4.7 — including laziness, safety overreach, and over-engineering on simple tasks — may have been effort miscalibration rather than model failure. He walks through the CLI interface and Claude.ai effort controls, and previews dynamic workflows, a new feature for tackling very large-scale, multi-step problems.
The video also references Anthropic’s published benchmarks on “misaligned behavior,” where Opus 4.8 scores roughly half of 4.7, and includes community-sourced complaints about 4.7 (including perceived stubbornness and inconsistent completion reporting) as context for why the update matters. Herk recommends users experiment with effort settings before attributing poor output to the model itself.
📺 Source: Nate Herk | AI Automation · Published May 28, 2026
🏷️ Format: Review
