Running OpenClaw 24/7 with a frontier model for every task gets expensive. OpenClaw supports multi-model routing — assign different models to different tasks based on complexity.
The Strategy
| Task | Model | Why |
|---|---|---|
| Cron jobs | Haiku | Simple, repetitive |
| Conversations | Sonnet 4.6 | Best balance |
| Complex reasoning | Opus / GPT-5 | When truly needed |
| Local tasks | Ollama (Qwen) | Free |
Configuration
agents:
default:
primaryModel: anthropic/claude-sonnet-4-6
cron:
primaryModel: anthropic/claude-haiku-4-5
More Cost-Saving Tips
- Enable prompt caching for repeated system prompts.
- Reduce maxTokens (8192 instead of 16384).
- Use event-driven triggers over fixed polling.
- Batch notifications into summaries.
- Run Ollama locally for simple queries at zero cost.
Smart routing can cut costs 50-70% without sacrificing quality. Start by moving cron jobs to Haiku.
More at AiX Society.


