Understand AI costs before you spend
A low advertised price does not tell you what a finished assignment, illustration or video will cost. Start with the work you need to complete, then identify the units charged, the number of attempts and any extra steps. This guide uses invented prices to teach the arithmetic. The examples are not offers or quotes, and real providers may measure or charge for work differently.
Published October 16, 2026 · Benny Labs Foundation · Free to read, no registration
AI text cost calculator
Enter your estimated token counts and rates in US dollars per one million tokens. The starting prices are fictional teaching examples, not current vendor prices. This estimates token charges only; tools, subscriptions, taxes and other fees are excluded.
Example estimate: $3.00 USD
Input: $1.00 + output: $2.00
The calculator runs in your browser. It does not upload your entries or require an account or API key. For a cost below one cent, the result shows additional decimal places.
Know what one million tokens means
Text models often price input and output separately, using a rate per one million tokens, written as 1M tokens. Tokens are pieces of text or other encoded information; they are not interchangeable with words or characters. Token counts vary with the model, language and content. Use the selected model’s tokenizer or usage report when available rather than assuming a fixed words-to-tokens conversion.
Input is material the model receives. It can include your instructions, attached text, conversation history and information returned by tools. Output is material the model generates. Some services also bill other processing, such as reasoning or tool use. Check which categories appear on the actual bill. A long conversation can become more expensive when earlier messages are sent again.
Calculate each charge before adding it
For a token-based estimate, divide input tokens by 1,000,000 and multiply by the input rate. Do the same for output tokens and the output rate, then add the results. Keep the currency and billing period consistent. Treat an estimated output length as a budget assumption; the model may produce more or less than you expect.
Worked example 1 — Hypothetical input costs $1 per 1M tokens and output costs $4 per 1M tokens. For 1M input tokens, the input charge is 1 × $1 = $1. For 500,000 output tokens, the output charge is 0.5 × $4 = $2. The estimated total is $3, before any separate fees or taxes.
Images and video use different units
Image generation may be priced per image, by resolution or quality, or through another metered unit. Video may be priced per generated second, per clip or by credits. Confirm whether the rate applies to every generated candidate, not just the result you download. Editing, upscaling, audio and extending a clip may introduce additional charges.
Worked example 2 — Suppose an invented image service charges $0.04 per generated image and $0.02 per upscale. You need 12 finished images and budget three candidates for each: 12 × 3 = 36 generations. Generation costs 36 × $0.04 = $1.44. Upscaling the 12 chosen images costs 12 × $0.02 = $0.24. The estimated total is $1.68, before any other charges.
Budget for attempts, caching and subscriptions
Count attempts rather than only finished results. Prompts may need revision, and successful requests can still produce unusable material. Check how failed requests, cancellations and retries are charged; do not assume every failure is free. Run a small representative sample, record its cost and acceptance rate, then leave room in your budget for variation.
A cached-input discount may apply only to eligible repeated content, under conditions such as minimum size or cache lifetime. Do not apply that discount to every input token. A subscription can also have message limits, credits, slower queues or restricted features. Check whether API usage is included or billed separately, and whether unused credits expire.
Compare the cost of acceptable work
Use the same task, input length, output requirements and quality threshold when comparing options. A cheaper generation that requires several retries or substantial editing may cost more to finish. For schoolwork, consider whether the result needs accurate citations or careful checking. For creative work, include resolution, consistency and any permissions required for your intended use.
A comparison table is a starting point. Prices can change, currencies may differ, and credits do not have a universal value. Promotional rates may expire. Availability, usage caps and the provider’s terms can matter more than a small difference in the headline rate. Check the provider’s current billing information before committing money.
Protect your information and keep a simple record
Before uploading class materials, client work or personal information, check the service’s rules for storage, retention, training use and deletion. Follow your school’s or client’s requirements. Remove unnecessary names and sensitive details, and never include passwords or access keys. Price alone cannot tell you whether a service meets your privacy needs.
- Write down the task, estimated units, rates, attempts and expected total before starting.
- Set a spending limit or usage alert where available; check how quickly it takes effect.
- After a small trial, compare the estimate with actual usage and revise your budget.
- Record costs per accepted result, including extra processing, so your next comparison reflects the work you actually need.
Keep learning
Browse the free learning library or explore our public programs and research explanations.
Have a correction or an educational question? Contact the Foundation.