What is your AI actually costing you?
Paste a prompt, pick your models, and get an instant cost breakdown across OpenAI, Anthropic, and Google, then see how your spend stacks up against businesses that run their AI bills through Ramp, benchmarked against real Ramp platform data.
Price out a prompt across every major model.
Estimates use each provider's published per-token pricing as of July 2026. Token counts are approximate: real tokenizers vary by 10–15% depending on content.
Show token breakdown
Customize
This is a separate assumption from the prompt above — it doesn't scale with what you paste in, since response length depends on the task, not the prompt's length. A one-line answer is roughly 50–150 output tokens; a ~600-word memo is roughly 800–900; long-form writing or code often runs well past that. Pick a preset or enter your own.
| Model | Input | Output | Per request | Per month |
|---|
Monthly cost by model
At your current volume, sorted from cheapest to most expensive.
What is a token, and why does it decide your AI bill?
Every prompt and response is billed in tokens, not words or characters. A token calculator breaks your text into the same kind of chunks a model's tokenizer uses, so you can see the input cost before you send a request, and the output cost you're on the hook for after it responds.
How to use this calculator
- Paste a prompt, document, or code snippet into the box above.
- Watch the live token breakdown and per-model cost update as you type.
- Set requests per day to see realistic monthly spend, not just one call.
- Pick a primary model to compare against the cheapest option for that same prompt.
Input tokens vs. output tokens
- Input tokens
- What you send: the prompt, system message, and any context.
- Output tokens
- What the model generates back.
Providers price them differently, which is why a short prompt asking for a long response can cost more than a long prompt asking for a short one.
How your estimate compares to real businesses.
Ramp processes AI vendor spend for tens of thousands of businesses. Here's what real businesses actually pay, not just who's winning the AI market.
Effective AI token costs across the TSM cohort.
Median cost per 1M tokens observed among Ramp's Token Spend Management customers (API-key-connected, Anthropic + OpenAI + Cursor), Feb–Jul 2026. An observed rate within this cohort, not population-wide list pricing.
Overall adoption figure corroborated by Ramp's public AI Index. Token-pricing figures reflect an observed median within Ramp's Token Spend Management cohort (businesses with a connected Anthropic, OpenAI, or Cursor API key), not population-wide list pricing; medians used throughout rather than averages, since a small number of large spenders would otherwise dominate the numbers.
Curious about your actual AI spend?
See where your team's AI budget goes, broken down by provider, model, and user, connected directly to your own accounts.
Try AI Token Spend ManagementToken calculator FAQ
Yes. This one prices input and output tokens across OpenAI, Anthropic, and Google at once, using each provider's current published rate, and benchmarks your estimate against real Ramp platform spend data below.
A tokenizer breaks text into sub-word chunks: roughly 4 characters or 0.75 words per token for English. This calculator estimates your input tokens that way as you type, then multiplies input and output token counts by each model's per-token rate to get a cost.
It depends entirely on the model. Across the models in this calculator, 1,000 input tokens costs between $0.0002 and $0.03, and 1,000 output tokens costs between $0.0012 and $0.18, a 150x spread from cheapest to most expensive. Enter your own token count above to see the exact cost per model.
Again, it's model-dependent: $1 buys anywhere from about 33,000 to 5 million input tokens, or 5,500 to 833,000 output tokens, depending which model you pick. The cheapest and most expensive models in this calculator differ by roughly 150x.
About 750 words for English text. The ratio holds roughly steady up to larger counts: 1,000,000 tokens is on the order of 750,000 words, though code and non-English text typically run fewer words per token.
Not for a modern model. 3,000 tokens is roughly 2,250 words, a few pages of text, while most current models support context windows from 128K up to 1M+ tokens. It's a meaningful chunk of a single prompt, but nowhere near a model's limit.
See every token. Control every dollar.
Connect OpenAI, Anthropic, and Gemini directly via your provider API keys to see spend by provider, model, key, and user in one dashboard. Free to get started, no Ramp card required.