Free Tool
LLM Token Cost Calculator
How much do GPT, Claude, and other LLMs cost per token? Enter your usage to see cost per request, per month, and per year — and compare every model side by side.
- Instant calculation
- Runs in your browser
- Compare all models
Representative list prices (updated —) — providers change rates often, so verify on the official pricing page. Token counts vary by content.
No spam — we only use this to send your breakdown and follow up about your project.
The same usage, priced across every model
Cheapest to most premium — at your current input/output/volume.
Turn per-token pricing into a monthly bill
AI pricing looks cheap per token and stops looking cheap around the second month. This calculator turns a per-million-token price into a monthly figure you can actually put in a budget.
Estimates go wrong because almost nobody gets volume right on the first attempt. A workflow that felt small in testing runs ten thousand times in production, and the bill arrives without warning.
The thing most people get wrong
Output tokens cost roughly five times what input tokens cost on every major model. That single fact changes which model makes sense.
- A task that reads a long document and writes a short summary is cheap on any tier.
- A task that generates long reports is where the tier choice hits hardest — and it is the one people assume will be similar.
- Estimate input and output separately. Treating them as one number produces an answer that is wrong by a factor of several.
How to estimate your volume
Volume is the number that breaks budgets, not price. Work it out deliberately rather than from the feeling you had while testing.
- Roughly 750 words per 1,000 tokens — the Token Counter gives you an exact figure for a real sample.
- Retries, failed runs and testing all consume tokens that nobody budgets for.
- The same workflow in testing and in production is the same price per token and a completely different bill.
- 1Count the runsHow many times a day does this workflow fire?
- 2Estimate tokens per runAbout 750 words per 1,000 tokens.
- 3Split input and outputThey price differently.
- 4Add a marginRetries, failed runs and testing.
Two discounts that change the answer
Neither needs a different model. Batching costs nothing but patience, and caching pays off on exactly the workloads that look most expensive — the ones repeating a long system prompt on every single call.
- Batch processing trades immediacy for roughly half the price, on work that does not need an answer this second.
- Prompt caching cuts the cost of the repeated part of a prompt dramatically once the same context is sent again and again.
- Check both before concluding a workflow is too expensive to ship.
The number that actually matters
Not price per token. Cost per completed task. A cheaper model that fails one time in three is more expensive than a pricier one that finishes, once you count the retries and the human cleanup.
- Run both on the same job, with the same prompt and the same definition of done, then compare totals.
- The AI Model Comparison Table lists the per-model pricing to start from.
- Picking the cheapest model is as naive as always picking the best one. Neither is a strategy.
Four numbers worth keeping
Why Parix.ai gives this away
We build AI integrations for a living, which means we size these bills before we build. These tools are the same checks we run for clients, put on a public page. They are how people find us — which is the whole reason there is no paywall and no sign-up on any of them.
Estimate my billWho uses it
Engineering leads sizing a feature before it ships. Founders working out whether an AI product has a margin. Finance teams handed an API invoice they were not expecting.
Pair it with
The API Cost Calculator for the non-AI services in the same workflow, and the Token Counter for an exact token figure on a real sample.
Checked against a real build
Our AI integration team will sanity-check an estimate against what the workflow actually costs once it is running.