Savings methodology

The savings figures on this site are measured results from specific example workloads run through IQ Routing against a single-model baseline. Your own savings scale with your model mix, cache hit rate, and the shape of your agent loops, so these examples are not a guarantee of the result any single workload will see. This page describes where the numbers come from so you can judge how they map to your own usage.

Cut agent loop costs 40 to 60 percent

The headline range describes the spread we see across agent workloads when the resolver picks the cheapest model that holds the quality bar on each step instead of pinning every step on one top-tier reasoning model. The size of the cut depends on how many steps in the loop genuinely need a frontier model. A loop that is mostly retrieval, tool calls, and formatting moves toward the high end of the range. A loop where almost every step is hard reasoning moves toward the low end, because there is less room to route down. The range is a band; where your workload lands within it depends on its step composition.

The 49 percent before and after example

The before and after example on the home page is a single five-step agent loop: planning, retrieval, a tool call, synthesis, and verification. In the Before trace every step runs on the same top-tier reasoning model at high thinking effort, which totals $1.84 for the loop. In the After trace IQ Routing resolves each step to the cheapest model that clears the quality bar for that step, so planning and synthesis stay on the heavy reasoning model while retrieval, the tool call, and verification drop to cheaper variants. That total is $0.94, which is 49 percent less than the baseline.

This is one representative loop chosen to show the mechanism. The per-step prices reflect a specific mix of providers and step types. Your own savings scale with these factors: a loop with a different shape, a different provider mix, or a higher share of reasoning-heavy steps will land at a different number.

Why results vary by workload

Three factors drive most of the variance. The first is model mix: the more of your traffic that can run on a cheaper model without losing quality, the larger the cut. The second is cache hit rate: repeated or near-repeated prompts served from the semantic cache cost nothing on the provider side, so workloads with high reuse save more, while first-of-a-kind prompts save less. The third is step composition: agent loops with many light steps (retrieval, tool calls, short formatting) leave more headroom to route down than loops dominated by hard reasoning. Your own savings depend on how these three land for your traffic.

How to read these numbers

The figures on this site are worked examples that show what the routing and caching mechanisms do. The dashboard reports your own routed cost against a single-model baseline, so you see the exact figure for your traffic. The way to size the savings for your workload is to run it through IQ Routing and read the measured result, which scales with your model mix, cache hit rate, and step composition.