Two token prices
Every model charges per million tokens, split into input (your prompt) and output (its reply). Output is almost always the pricier half, often several times the input rate.
What will your AI feature actually cost?
Enter your tokens per request and requests per month. This LLM cost calculator turns them into a real monthly bill in pounds, and shows Claude, GPT and Gemini side by side on your own numbers, so you can price a feature before you ship it.
Estimated monthly cost
– / month
– per request
– per 1,000 requests
Every model, your numbers
Estimated monthly cost at your input, output and request counts.
| Model | $/1M in | $/1M out | Est. £/mo |
|---|
Prices verified 12 July 2026. Standard per-token list rates, USD per 1,000,000 tokens, converted at $1 = £0.749 (Bank of England spot, early July 2026). These exclude batch, cached-input and volume discounts, and model line-ups change often, so always confirm before you rely on a figure. Sources: Anthropic, OpenAI, Google.
How the cost is worked out. Token by token.
LLM pricing looks fiddly but it's four inputs and one sum. Once you see it, you can price any feature on the back of an envelope.
The per-request sum
Input tokens over a million times the input price, plus output tokens over a million times the output price. That's the cost of a single call, usually a fraction of a penny.
Times your volume
Multiply by requests per month and convert to pounds. Small per-call costs turn into a real monthly bill at scale, which is exactly what the table shows across every model.
The cheapest model isn't always the cheapest
A lower per-token price can still cost more if it needs longer prompts, more retries or bigger replies to do the job. The honest comparison is your real workload, your token counts and your volume, priced across every model at once. That's what this tool does, so you're choosing on the numbers that matter to your bill, not the headline rate.
- Your real monthly bill in pounds, not a per-token abstraction.
- The per-request and per-1,000-request cost, for unit economics.
- Every model priced on the same inputs, side by side.
- Where output length quietly dominates the total.
You've priced it. Now go build the thing.
A cost you can live with is the green light. The rest is shipping, and there's a tool, a course and a room full of builders for exactly that.
Validate it, de-slop it, get it found
Free tools to validate the idea, de-slop the design, and check it's actually found by AI. Built by a marketer who ships real software with Claude, not another anonymous SaaS.
Questions people ask.
You pay per token, split into input (the prompt you send) and output (the reply). For one request: cost equals input tokens divided by a million times the model's input price, plus output tokens divided by a million times its output price. Multiply by your requests per month for the monthly bill. This calculator does all of that and shows the result in pounds.
Input tokens are everything you send the model: the system prompt, the conversation so far, and the user's message. Output tokens are what the model writes back. They're priced separately and output almost always costs more, often three to six times the input rate, so long replies are what drive a bill up fastest.
A rough rule of thumb for English is that one token is about four characters, so roughly 100 tokens is about 75 words. Code, punctuation and other languages tokenise differently. For a real bill, count tokens with the provider's own tokenizer or token-counting endpoint rather than estimating from word count.
The prices in this tool are dated and shown on the page, and they're the standard per-token list rates, not batch or cached-input discounts. Model line-ups and prices move often, so always confirm against the provider's own pricing page before you rely on a figure for budgeting. Links to each provider's pricing sit right under the calculator.
Generating text is more compute-intensive than reading it: the model produces output one token at a time, each one conditioned on everything before it, while input can be processed in parallel. That's why every provider charges more for output, and why trimming reply length is usually the biggest lever on cost.
Yes, completely free with no card and no sign-up. Everything runs in your browser, the comparison updates live as you change the inputs, and each scenario has its own shareable link.
The cost is fine. Go and ship it.
You know what your AI feature costs to run. The only thing left is to build it, so start with the free tools that take you from idea to shipped.
Explore the builder tools →
Validate the idea, de-slop the design, check it's found by AI. All free, all built live with Claude.