AI model cost calculator.

Compare GPT-6.2 Astra and Sol token costs for your workload. Adjust usage and rates, then copy the estimate. Free, no signup.

Your expected usage

Start with an example, then use the token counts from your API usage. Count every call in a multistep agent task.

Edit rates in USD per million tokens
GPT-6.2 Astra
GPT-6.2 Sol
Defaults checked November 17, 2026.

Estimated monthly token cost

GPT-6.2 Astra

$8.00

$0.08 per call · 100 calls

Uncached input / call
$0.04
Cached input / call
$0.00
Output / call
$0.04

GPT-6.2 Sol

$2.00

$0.02 per call · 100 calls

Uncached input / call
$0.01
Cached input / call
$0.00
Output / call
$0.01

Using published standard API rates. This compares token spend for identical usage, not model quality or cost per successful task.

Give your chosen model traffic and funnel evidence with Humblytics, then review the change it proposes. See the AI CRO workflow →

What does an AI workflow cost?

Multiply each billed token category by its price, add those costs, then multiply by the number of API calls. One agent task can make several calls, so one task is not always one request.

The example uses 5,000 input tokens, no cache hits and 1,000 output tokens per call. At the published rates, 100 identical calls cost $8 in Astra token usage or $2 in Sol token usage. This is arithmetic for an illustrative workload, not an observed task benchmark.

USD per million tokens

Standard OpenAI API rates checked November 17, 2026
ModelInputCachedOutput
GPT-6.2 Astra$8.00$0.40$40.00
GPT-6.2 Sol$2.00$0.10$10.00

Standard rates checked November 17, 2026. Source: OpenAI's GPT-6.2 announcement. Edit the defaults if your rate differs.

How to estimate API token cost

Cost = uncached input + cached input + output

For each category, cost equals tokens × USD per million tokens ÷ 1,000,000. Cached input is subtracted from total input before the normal input rate is applied, so it is never counted twice.

Does this include a ChatGPT or Devin subscription?

No. This estimates API token charges only. It does not calculate ChatGPT plans, Codex subscription allowances, Devin credits, tool charges, hosting, taxes or other services. Use your provider's usage and billing records for an actual bill.

How many tokens should I enter?

Use representative billed usage from your API responses or usage dashboard. Include the conversation and other context counted as input, and the billed output count. Repeat this for all calls in an agent workflow. The defaults are an example, not a typical workload claim.

How do cached input tokens affect cost?

Enter cache hits as a subset of total input. With 5,000 input tokens and 1,000 cached tokens, the calculator charges 4,000 at the input rate and 1,000 at the cached rate. Enter zero if you do not know whether your requests qualify for caching.

Does the cheaper estimate identify the better model?

No. The comparison holds token usage equal. Models can use different numbers of tokens, need different numbers of attempts and produce different results. Evaluate the same task, check its correctness and measure actual usage before deciding which model is better value.

Can I use a custom API rate?

Yes. Open Edit rates and replace the input, cached input and output prices. Results are labelled as custom rates. Reset example restores the published standard defaults. The calculator does not apply automatic batch, service-tier or other discounts.

Budget the task, then check the answer

For website work, compare how models preserve dates and denominators, propose a testable change and explain an inconclusive result. Use the AI model guide for CRO and A/B testing to design that evaluation.

Start with the AI CRO brief generator to give each model the same assignment. For recurring work, the Devin Bots conversion workflow explains how to keep the job bounded and review changes before launch.

Free to start · 5,000 data points, no card

Give your agent real website evidence.

Connect Humblytics so your agent can read traffic and funnels, draft a test and bring the next step back for review.

No card. Team plans from $79 a month. Book a demo