Pareto Optimization

The Trade-off

Every request faces a choice: pay more for better quality, or save money with a cheaper model. The Pareto optimizer finds the sweet spot based on your preferences.

Four Modes

Eco

objective = 0.05 * quality + 0.95 * cost

Minimizes energy consumption. Routes almost everything to the smallest, fastest models. Reduces compute footprint. Good for high-volume workloads where quality can take a backseat.

Cheap

objective = 0.15 * quality + 0.85 * cost

Prioritizes cost savings. Sends most requests to cheap providers. Good for high-volume, low-stakes usage.

Balanced

objective = 0.55 * quality + 0.45 * cost

The default. Balances quality and cost equally. Works well for most use cases.

Accurate

objective = 0.85 * quality + 0.15 * cost

Prioritizes quality. Routes complex prompts to the best available models. Use this for critical tasks.

Changing Modes

Edit ~/.config/fastlane/settings.toml:

[general]
mode = "eco"  # eco, cheap, balanced, or accurate

Or via the API:

curl -X POST http://127.0.0.1:8787/v1/settings \
  -H "Content-Type: application/json" \
  -d '{"general": {"mode": "eco"}, "byok": {}, "affiliate": {}}'

Restart the proxy for changes to take effect.

Visual Example

For a complex coding prompt in balanced mode:

Model	Quality	Cost	Objective
gemma2-9b-it	0.88	0.93	0.91
claude-sonnet-4	1.00	0.02	0.55
gpt-4o	0.95	0.17	0.58

The optimizer picks the best value across the quality-cost spectrum.