Understanding Cost Avoidance
Understanding Cost Avoidance
Savings represent pure Cost Avoidance. They are calculated deterministically on a per-request basis against the flagship 2026gpt-5.5model.If a routed model costs less than the baseline, the difference is added to your savings.
If you intentionally route a task to a higher-cost premium SKU, the savings for that specific request is strictly$0.00.Your cumulative savings track how much total money you avoided spending across all efficient routing decisions. Negative savings are mathematically floored to zero to ensure your dashboard accurately reflects pure saved capital without penalizing intentional premium model usage.
How it is computed in the platform
For each metered request, the gateway records a baseline cost using gpt-5.5 list economics ($5.00 per million input tokens and $30.00 per million output tokens) applied to the same token counts as the routed call. Your wholesale spend for the actual model is stored separately.
Cost avoidance for the period adds up, request by request, how much cheaper the model you actually ran was compared with the flagship baseline: if the baseline would have cost more, only that positive difference counts toward savings for that request. If the route you chose was more expensive than—or the same cost as—the baseline, that request adds nothing to the savings total (we never subtract, so choosing a premium model never drags the cumulative number down).
Platform fees are not part of this comparison—the metric is pure compute against the flagship baseline.