Benchmarks
Reproducible cost and latency benchmarks for ClawRouter routing profiles.
Cost Comparison
| Profile | Blended Cost | Savings vs Claude Opus |
|---|---|---|
| Eco | ~$0.10/M tokens | 98% |
| Auto (default) | ~$2.05/M tokens | 92% |
| Premium | ~$25/M tokens | 0% |
How It Works
ClawRouter's 14-dimension scorer classifies requests by complexity:
- Simple queries → free/cheap models ($0/M)
- Medium tasks → mid-tier ($0.10-$1.50/M)
- Complex reasoning → premium ($2-$12/M)
Reproduce
git clone https://github.com/Clawland-AI/OpenclawBox.git
cd OpenclawBox
python benchmarks/run_bench.py
Disclaimer: Up to 90% cost reduction in certain workloads. Results vary.