LLM API
cost calculator
Paste a sample prompt or type token counts, set your volume, and see what a month costs on Claude, GPT, Gemini, Llama and DeepSeek - side by side, with every price editable.
What will a month of calls cost?
Token estimate = characters / 4. Real tokenizers differ by model and language, typically by 10-15% - good for budgeting, not for invoices.
| Model | Input $/1M | Output $/1M | Cost / month | Compare | If it 10x's |
|---|
When your AI builds a website, Yetty puts it live and stays connected over MCP - your AI keeps editing, publishing and reading leads.
How LLM billing
actually works
Two meters, not one
Input and output are billed separately, and output usually costs about 5x more per token. A chatty model on a cheap plan can out-bill a terse model on an expensive one.
You resend the whole chat
The API is stateless: every turn resends the full history as input. A 20-turn conversation pays for turn 1 twenty times - which is why long chats get expensive fast.
Caching is the big lever
A stable prompt prefix (system prompt, docs, tools) can be cached; providers then bill cached input at a heavy discount - often around 90% off. Toggle it above to see the effect.
Model size is a 25x range
Between a small model and a flagship the same call can differ by 25x in price. Route simple calls (classify, extract, reformat) to the small one and keep the flagship for hard work.
Tokens are not words
English averages ~4 characters per token, but code, JSON and non-Latin languages tokenize heavier. The same "short" prompt in Hebrew or Japanese can cost noticeably more.
Success is the danger
Costs scale linearly with usage - the 10x column above is the "our feature took off" scenario. Budget for it before launch, not in the invoice email after.
People also ask
How much does the Claude API cost per million tokens?
As of mid-2026 the headline rates are about $5 per million input tokens and $25 per million output tokens for Claude Opus, $3/$15 for Claude Sonnet and $1/$5 for Claude Haiku. Providers change prices and run promotions often, which is exactly why every price cell in this calculator is editable - treat the defaults as starting points and paste in the current numbers from your provider's pricing page.
How do I estimate how many tokens my text is?
A quick rule of thumb for English is one token per 4 characters, or roughly 750 words per 1,000 tokens. Real tokenizers differ by model and language - code, non-Latin scripts and unusual formatting can shift the count by 10-15% or more - so use the estimate for budgeting and the provider's own token counter when you need an exact bill.
How can I reduce my LLM API costs?
The three levers that matter most: use a smaller model for simple calls (classification and extraction rarely need a flagship), cache your repeated prompt prefix (providers discount cached input heavily - often around 90%), and trim what you send - shorter system prompts and summarised history cut the input bill on every single request. Batch APIs typically add another ~50% off for jobs that can wait.