AI Cost Calculators
AI Model Routing Savings Calculator
Compare an all-premium AI workflow with a routed low-cost and premium model strategy, including fallbacks, retries, gateway fees, setup, and unresolved requests.
Enter Your Model Routing Plan
Compare an all-premium baseline with a lower-cost routed workflow.
Models and Monthly Workload
Routing, Quality, and Fallback
Current Model Prices
Router, Platform, and Setup Costs
Estimated routing flow
Low-cost routed requests: 70,000
Direct premium requests: 30,000
Low-cost requests passed: 61,600
Premium fallback requests: 8,400
Estimated unresolved requests: 0
Estimated completion rate: 100%
Model Routing Cost and Savings
The result compares the routed planning cost with an all-premium baseline using the same workload and retry overhead.
Routed monthly planning cost
All-premium baseline
—
Monthly planning saving
—
Cost per resolved request
—
Low-Cost Model usage
72,800 calls including retry overhead
—
Direct Premium Model usage
31,200 calls routed directly to the premium path
—
Premium Model fallback
8,736 fallback calls including retry overhead
—
Router or classifier
100,000 routing decisions
—
Routing platform fee
0% of model spend
—
Fixed monthly routing cost
Gateway, observability, evaluation, or routing platform
—
Amortised implementation
$0.00 spread across 12 months
—
Routed monthly operating cost: —
Monthly operating saving: Enter current model prices
First-year comparison: Enter current model prices
Approximate routing-share break-even: Enter current model prices
Implementation payback: Enter current model prices
Cost per attempted request: —
Estimated unresolved requests: 0
Budget status: Add a budget to compare
* Important: This calculator stores no model or gateway price. Enter the current official rates for the exact low-cost model, premium model, router, and platform being considered. The quality pass rate and fallback coverage should come from evaluations or production data. Blank optional cost fields are treated as zero.
A model router can send routine requests to a lower-cost model and reserve a stronger model for complex work. The saving depends on routing share, quality pass rate, fallback behaviour, retries, and the cost of running the router itself.
Comparing Routing With an All-Premium Baseline
Enter monthly request volume, token usage, and the current input and output prices for a low-cost model and a premium model. The baseline assumes every request uses the premium model.
The routed workflow sends the selected traffic share to the low-cost model. Requests that do not pass the quality check can be escalated to the premium model according to the fallback coverage entered.
The result separates low-cost-model spend, direct premium spend, premium fallback spend, gateway charges, fixed costs, and amortised setup.
Including Quality and Fallback Behaviour
A routing plan is not complete without a quality estimate. The low-cost model pass rate should come from evaluations or production logs for the exact tasks being routed.
Fallback coverage controls how many failed low-cost attempts are retried on the premium model. Full fallback coverage can protect completion rate but increases premium usage.
The calculator shows the estimated unresolved request count when failed low-cost requests are not fully escalated.
Accounting for Router and Gateway Costs
Routing can be implemented with deterministic rules, a classifier, a separate model call, or a managed gateway. Add the combined router cost per 1,000 requests, any percentage platform fee, and fixed monthly gateway cost.
One-time implementation and evaluation work can be spread across a chosen number of months. This produces a planning cost that is more realistic than looking only at inference tokens.
Using Routing Safely
Route only task groups that have clear evaluation results. High-risk, ambiguous, regulated, or difficult requests may need to stay on the premium path.
Track model choice, cost, latency, fallback rate, accepted output, and unresolved requests. Recheck thresholds when prompts, models, providers, or traffic patterns change.
Amazon Bedrock documents intelligent prompt routing as a way to balance response quality and cost within supported model families. OpenRouter also supports model and provider routing, fallbacks, budgets, and spend controls.
Practical Decisions This Tool Supports
- Estimate savings before building a model router.
- Compare all-premium and tiered-model inference costs.
- Measure the premium fallback cost of quality failures.
- Include gateway, platform, retry, and implementation costs.
- Estimate unresolved requests under partial fallback.
- Find the minimum low-cost routing share needed to break even.
- Calculate cost per request and cost per resolved request.
- Check the routed workflow against a monthly budget.
Costs and Risks Outside the Estimate
The result does not automatically value answer quality, latency, safety, customer satisfaction, regulatory risk, or provider reliability. It also excludes taxes, data transfer, observability, support, evaluation labour, and unexpected provider changes unless entered.
Average token values can hide expensive long requests. Review percentile-level production data before setting routing thresholds or customer pricing.
Frequently Asked Questions
What is AI model routing?
Model routing sends different requests to different models. Simple or predictable tasks may go to a lower-cost model, while difficult requests, high-risk work, or failed attempts are sent to a stronger premium model.
Why does the calculator include a low-cost model pass rate?
Not every request routed to a smaller model will meet the required quality level. The pass rate estimates how many routed requests finish successfully without using the premium fallback.
What is fallback coverage?
Fallback coverage is the share of failed low-cost-model requests that are retried on the premium model. A lower value can save money but may leave more requests unresolved.
Why are all model prices blank?
Routing can combine any providers or private agreements. Blank fields prevent example rates from looking like current official prices. Enter the exact live rates for the models and gateway being considered.
How is the break-even routing share calculated?
The tool scans possible low-cost routing shares and finds the smallest share where the routed monthly planning cost is no higher than the all-premium baseline.
Does lower cost mean the routing plan is better?
No. Quality, latency, safety, regional availability, provider reliability, fallback behaviour, observability, and unresolved requests should be reviewed alongside cost.