AI Token Calculator
Estimate input tokens from pasted text or character count, reserve output tokens, check entered context usage, and estimate cost from custom per-million rates.
Estimated input tokens
18tokens
0.80% of the entered context including planned output
- Planned total
- 1,018
- Cost per request
- $0.01
- Total estimated cost
- $25.09
Token and cost plan
- Characters counted
- 72
- Words counted
- 13
- Estimated input tokens
- 18
- Planned output tokens
- 1,000
- Context remaining
- 126,982
- Input cost per request
- $0.000036
- Output cost per request
- $0.01
- 2,500 requests
- $25.09
Guidance checked 2026-08-21
Four characters per token is an editable English rule of thumb, not a model-exact count. Use the target provider's counter before billing or context decisions.
What this calculator assumes
- Pasted text is counted locally in this browser and is not sent to an AI provider.
- The default four-character ratio is an English planning estimate; language, code, spaces, punctuation, and tokenizer version can change the count.
- Context window and prices are your inputs, not current provider presets.
- Tools, media, system overhead, cache behavior, reasoning tokens, retries, tiers, taxes, and minimum charges are excluded unless reflected in your inputs.
Formulas, assumptions, and rounding are documented in our calculator methodology.
Built and maintained by Will Henschell, Data Scientist. Editorial policy.
The Transparent Token Estimate
Estimated input tokens equal character count divided by the editable characters-per-token ratio, rounded up to a whole token. The default ratio of four follows OpenAI's current English rule of thumb. The page shows the actual local character and word counts so the estimate can be audited instead of presenting tokenizer-like precision it does not have.
Plan Input and Output Together
Context capacity normally has to accommodate both the supplied input and generated output. Add a realistic output allowance, enter the target model's context window, and review total usage rather than checking the prompt alone. A result above 100% is a planning warning to shorten, split, summarize, retrieve selectively, or choose an appropriate model.
Use Current Custom Rates for Cost
Input and output often have different prices, so the calculator prices them separately. No provider preset is embedded here. That keeps the page focused on text-to-token planning and avoids duplicating the LLM API Cost Calculator, which compares broader token classes and provider presets. Use account-specific rates for the closest forecast.
Exact Counting Requires the Target Model
OpenAI notes that tokenization varies by model and encoding. Anthropic's counting endpoint takes a selected model and can still differ slightly from message creation, while Google's countTokens runs the selected model's tokenizer on supplied content. For bills and hard context limits, use the same model and full structured request you will actually send.
Official Source Review — August 21, 2026
The English four-character approximation was rechecked against the official OpenAI tokenizer. Exact-count limitations and model-specific alternatives were rechecked against current OpenAI, Anthropic, and Google documentation. No provider price or context-limit constant is shipped; every potentially changing commercial or capacity value is an explicit user input.
Frequently Asked Questions
- OpenAI's current English rule of thumb is about four characters per token, or about three-quarters of a word per token. It is only an approximation. The calculator exposes the character ratio so you can change it for a measured workload.
- No. This calculator counts characters and words locally in the browser and sends no pasted text to an AI provider. The site still cannot produce an exact provider/model tokenization without running that model's tokenizer.
- Providers and models use different tokenizers, and spaces, punctuation, partial words, code, language, system messages, tool definitions, images, PDFs, and other structure can change token usage. Anthropic and Google provide model-specific counting endpoints, while OpenAI recommends its tokenizer or tiktoken for exact target-model work.
- Estimated input tokens and planned output tokens are added, then divided by the context window you enter. The tool shows remaining tokens or an overage. It does not supply a model limit because limits can differ by model version, tier, modality, and provider.
- Estimated input and planned output tokens are priced separately per million tokens, then added for one request and multiplied by request count. Enter current rates from the intended model and account. Caching, reasoning, tools, retries, tiers, media, storage, taxes, and minimum charges can add cost.