You’re Using Opus 5 Wrong. Here’s The Fix

An analytical overview of Anthropic’s Claude Opus 5 release, highlighting practical usage tips, cost-performance trade-offs, and security implications. The video evaluates reasoning modes, prompt patterns, model selection versus tasks, and comparisons with competitors like Fable 5. It also demonstrates Opus 5’s capabilities on technical tasks (3D animation and fluid simulation) and discusses vulnerability detection versus exploit generation.

– Never use max reasoning effort: Opus 5’s “max” mode increases cost (~14%) and can worsen results; use x-high for complex coding/data tasks and start with medium for general jobs.
– Drop verification prompts: Anthropic documents that verification instructions cause over-verification on Opus 5; removing them saves tokens, time, and context without quality loss.
– Prefer Opus 5 over Fable 5 by default: Opus 5 often matches or outperforms Fable 5 at much lower token cost; try Fable only after repeated failures.
– Match model to the task and mind security: use Haiku/Sonnet for lighter content and Opus 5 for technical work; Opus 5 finds vulnerabilities well, but exploitation risks remain.

Quotes:

Never use max reasoning effort.

Verification instructions cause over-verification on Claude Opus 5.

Quality doubled in 57 days — price stayed the same.