AI API Cost Calculator for GPT-5.6 and Claude 5

Use this AI API cost calculator to estimate a workload before moving traffic. It prices input and output tokens separately, includes retry volume, and keeps the model paired with a named route. You can edit either rate when the live Frugal Relay pricing catalog changes or when your account uses a different route.

Estimate AI API cost

Choose a model and route, then enter the token shape and request volume you expect.

Preset rates verified September 1, 2026

Input tokens per request Output tokens per request Requests per month Retry rate (%) Input price per 1M tokens (USD) Output price per 1M tokens (USD)

Cost per request $0.00

Expected calls with retries 0

Estimated monthly cost $0.00

Monthly input: $0.00 | Monthly output: $0.00

The preset values were verified against the public pricing catalog on September 1, 2026. They are planning inputs, not a billing guarantee. Your dashboard and usage logs remain the billing record.

AI API cost formula

For one token-priced request:

input cost  = input tokens / 1,000,000 x input route price
output cost = output tokens / 1,000,000 x output route price
request cost = input cost + output cost

The monthly estimate also includes retries:

expected calls = planned requests x (1 + retry rate / 100)
monthly cost = request cost x expected calls

This makes retry assumptions visible instead of hiding them inside a rounded budget. For agent workflows, count each model call in a tool loop as a request. A single user action can create several API calls.

How to estimate a production workload

1. Capture input and output tokens from a representative sample of real, redacted requests.
2. Use the model and named route available to the account that will run production traffic.
3. Enter the expected monthly request count and the retry rate observed in a controlled test.
4. Repeat the estimate for a typical request and a high-percentile request rather than relying on one average.
5. Compare models by cost per successful task, including retries and validation failures.

Token price alone cannot establish which model is the lowest-cost production choice. A less expensive call may still cost more per completed task if it needs more retries, longer repair prompts, or manual correction. Use the calculator beside a fixed evaluation set and a task-specific acceptance measure.

Current calculator presets

The tool includes these route-specific presets:

Model	Route	Input / output per 1M tokens
gpt-5.6-luna	Plus Account Route	$0.009 / $0.054
gpt-5.6-terra	Plus Account Route	$0.09 / $0.54
gpt-5.6-sol	Plus Account Route	$0.225 / $1.35
claude-haiku-4-5	Claude Sale Channel	$0.055 / $0.275
claude-sonnet-5	Claude Sale Channel	$0.11 / $0.55
claude-opus-5	Claude Sale Channel	$0.275 / $1.375

Rates and route eligibility can change. Check the full pricing guide and the live catalog before approving a production budget.

Cost controls after the estimate

An estimate is useful only when the deployment has an upper bound. Create separate API keys for development, evaluation, and production, then configure a daily or monthly limit for each key. Keep request timeouts and bounded retries in the application as well; a billing limit contains spend but does not decide which failures are safe to repeat.

For an existing OpenAI client, follow the OpenAI-compatible migration guide and change the configured base URL to https://frugalrelay.me/v1. For model names and endpoint types, use Supported Models.

FAQ

Does the calculator fetch my account or API key?

No. It runs in your browser and performs arithmetic on the values entered in the form. It does not require an API key and does not send a model request.

Why do input and output use different prices?

Token-priced models can have different input and completion rates. Output-heavy workflows can therefore have a different cost profile from classification or extraction jobs with short responses.

Should retries include agent tool loops?

Enter the actual number of planned model calls when possible. Use the retry field for repeated calls caused by failures or validation. If one workflow intentionally makes five model calls, count those five calls in the request volume instead of calling four of them retries.

Can I use a price that is not listed?

Yes. Choose Custom rates, then enter the input and output prices for the exact model and route you need to estimate.