Skip the coding benchmarks. The number that matters for your business is AutomationBench, which measures whether a model can carry a business task all the way to done instead of just starting it. Opus 5 passes roughly 1.5 times more of those tasks than the next best model at the same cost. Zapier's CEO ran it on a churn-prevention workflow. Previous models failed it. Opus 5 completed it end to end. Here's where to point it this week. 1. The retention sweep. Hand it your account or customer health data in whatever messy state it's in. Have it flag who's at risk, why, who owns the relationship, and what the next move is. This is the exact workflow Zapier tested and it's the highest dollar per hour thing on this list. 2. Work that lives inside software, not APIs. On the computer-use benchmark, Opus 5 beat Fable 5's best result at just over a third of the cost. That means agents that actually operate your CRM, your inbox, your dashboards. Not just ones that call a clean API endpoint. 3. Document-heavy work you've been avoiding. Box measured 11 percent better data analysis and 17 percent better due diligence versus Opus 4.8. Contracts, vendor reviews, proposal comparisons, financial packets. The pile you keep pushing to next week. 4. Run production automations on low effort. Opus 5 has an effort dial. Everyone will crank it to max. But at its lowest setting it still passes more business tasks than any other model on the market. Run your high-volume automations at low effort and you're paying pennies for output nobody else can match. 5. Delete your guardrail prompts. All those long verification instructions you wrote to stop the model doing something dumb? It now pushes back on flawed instructions on its own and splits big jobs into sub-agents. Those instructions are just tokens you're paying for. 6. Route, don't standardize. Opus 5 is your daily driver. Fable 5 for the longest autonomous runs. Haiku for high volume and low stakes. Picking one model for everything is the most expensive habit in AI right now.