You're paying $15/1M tokens for Claude Opus to summarize emails. Claude Sonnet does the same job for $3/1M tokens. That's an 80% savings you're leaving on the table—every single day.
The Problem: Default Model = Most Expensive Model
When you first set up OpenClaw, you probably configured Claude Opus (or GPT-4o) as your default model because it's the most capable. Makes sense for complex coding tasks. But here's the catch: every request goes through that model—including simple tasks like:
- Summarizing a short email
- Responding "nothing to do" to a heartbeat check
- Generating a one-line status update
- Answering a factual question with a known answer
These tasks don't need the most expensive model. A mid-tier model like Sonnet or GPT-4o-mini handles them identically—at a fraction of the cost.
How ClawBridge Detects This (Diagnostic A01)
ClawBridge's Cost Control Center scans your usage logs and flags when your primary model is one of the premium-tier options:
- Claude Opus → suggests Claude Sonnet (70% cheaper)
- GPT-4o → suggests GPT-4o-mini (80% cheaper)
- Claude 3.5 Sonnet → suggests Claude 3.5 Haiku (90% cheaper)
The diagnostic calculates your exact savings based on your actual token consumption over the past 7 days. No guessing—you see a dollar figure like "$8.40/week saved."
One-Tap Fix
Tap Apply in the Cost Control Center. ClawBridge:
- Backs up your current model configuration automatically.
- Writes the new model setting to your OpenClaw config.
- Logs the change with a timestamp for your records.
Changed your mind? Tap Undo to restore the original model instantly.
Trade-offs to Consider
- Quality on complex tasks: If your agent primarily does coding or multi-step reasoning, a cheaper model may produce lower quality. Consider this diagnostic mainly for agents doing routine work.
- Provider-specific features: Some models have unique capabilities (e.g., extended context windows). Make sure the suggested replacement supports your use case.
- Gradual testing: Try the cheaper model for a day before committing. Monitor your agent's output quality via Live Thoughts.
Real Numbers
| Scenario | Before (Opus) | After (Sonnet) | Monthly Savings |
|---|---|---|---|
| Light use (50K tokens/day) | $22.50/mo | $6.75/mo | $15.75 |
| Medium use (200K tokens/day) | $90/mo | $27/mo | $63 |
| Heavy use (1M tokens/day) | $450/mo | $135/mo | $315 |
FAQ
Q: Will my agent break if I switch models? A: No. OpenClaw's model configuration is a simple setting change. Your agent's skills, memory, and personality are unchanged.
Q: Can I switch back instantly? A: Yes. ClawBridge creates an automatic backup before every change. One tap to undo.
Q: What if I need Opus for some tasks but not others? A: Check out Multi-Model Routing (A10), which routes different task types to different models automatically.
ClawBridge is free and open source (MIT License) — install it in seconds, own it forever. Get ClawBridge Free →