Don't Use Premium Models as Default

Your default model is probably costing you 10x more than it needs to. Stop using premium models for everything.

Most Tasks Don't Need the Big Guns

The vast majority of agent tasks are simple — reading files, sending messages, checking calendars. These don't need frontier-model intelligence.

When Opus handles a task Sonnet does just as well, you're paying 15x more for zero benefit. My agent went from $50+/day to ~$8. The only change? A cheaper default model.

Set a Cheap Default

In your OpenClaw config, pick something economical:

• Claude Sonnet 4 — best balance of capability and cost
• Claude Haiku — even cheaper, handles routine tasks easily
• GPT-4o-mini — solid option at a fraction of GPT-4 prices

80% of tasks work perfectly with these models. File operations, basic automation, simple queries — none of these need a $15-per-million-token brain.

Route Expensive Tasks Explicitly

When you need the big brain — complex reasoning, nuanced writing, tricky code — invoke it deliberately:

• Use /model opus to switch for a specific conversation
• Configure certain skills to always use premium models
• Keep your baseline cheap, escalate when needed

Premium becomes a tool you reach for, not your default. Like a power drill — you grab it when you need it, not for every screw.

The Numbers

Before: daily costs like a hockey stick, some days hitting $80+. After: intensive days cost ~$15, average days under $10. Same productivity, fraction of the spend.

The math is simple — drop 80% of interactions from $0.50 to $0.03 and your average cost plummets. The occasional $2 Opus task barely registers.

You Won't Miss the Quality

For routine operations, Sonnet is indistinguishable from Opus. It reads files, runs commands, and formats output identically. Model differences only matter at the edges of reasoning and creativity.

Check your config. If it's Opus or GPT-4, switch it. Give Sonnet a week. Your wallet will thank you.