The published savings scenarios use documented workload presets. They are intentionally simple and explicit about what they assume.
Profiles
Each preset fixes the route mix and shows the reduction produced by that scenario—not a promise about future spend.
Cheap-first preset.
Max savings, highest variance.
Reduction
81%
Default mixed-lane preset.
Default tradeoff between savings and quality.
Reduction
63%
Higher quality-floor preset.
Higher quality floor, lower savings.
Reduction
42%
Inputs
Disclaimer
Savings vary with prompt mix, fallback rates, provider outages, token counts, and any routes you pin manually.
Comparison
| Profile | Reduction | Baseline | With Astrolabe |
|---|---|---|---|
Budget profile Max savings, most variance. | 81% | $1,000 | $190 |
Balanced profile Default tradeoff. | 63% | $1,000 | $370 |
Quality profile Higher floor, lower reduction. | 42% | $1,000 | $580 |
Route distribution
| Stack type | Budget | Balanced | Quality |
|---|---|---|---|
Budget lane summaries, lightweight extraction, classifier traffic | 70% | 45% | 25% |
Standard lane reasoning-heavy text work, longer synthesis, tool-aware chat | 22% | 40% | 45% |
Premium lane high-stakes actions, confirmations, or escalation retries | 8% | 15% | 30% |
Evidence
Snapshot date: May 1, 2026
Implementation
