Both families priced per million tokens, side by side — then paste your own prompt below to see what it actually costs on each one.
| Model | Input | Output | Total / request | Per 1,000 requests |
|---|
How this is estimated: token counts are approximated from your text (~4 characters or ~0.75 words per token — the same rule of thumb OpenAI's own docs use), then adjusted per model family. This is not an exact tokenizer, so treat numbers as a close estimate for budgeting — always confirm exact usage in your provider's dashboard before relying on it for billing. Pricing is manually verified against each provider's public pricing page and may change — check the source link for the latest rate before making a purchasing decision.
Both companies charge the same way: a price per million tokens going in, and a higher price per million tokens coming back out. Nothing is billed per request or per user, so the only things that move your bill are how much text you send and how much the model writes back.
| Model | Provider | Input/1M tokens | Output/1M tokens |
|---|---|---|---|
| GPT-5 | OpenAI | $3.00 | $15.00 |
| Claude Sonnet 4.5 | Anthropic | $3.00 | $15.00 |
| GPT-4o | OpenAI | $2.50 | $10.00 |
| Claude Opus 4 | Anthropic | $15.00 | $75.00 |
| Claude Haiku 3.5 | Anthropic | $0.80 | $4.00 |
| GPT-5 mini | OpenAI | $0.40 | $1.60 |
| GPT-4o mini | OpenAI | $0.15 | $0.60 |
Rates are the same ones the calculator on this page uses, checked September 2026. Providers change prices without much notice — the calculator is always the figure to trust.
GPT-5 and Claude Sonnet 4.5 both sit at $3 in and $15 out. For the models most teams actually build on, there is no cost argument between the two — pick whichever handles your task better, because the bill will come out the same.
At the top, Claude Opus 4 costs $15 in and $75 out — five times GPT-5 on both sides. It is the most expensive mainstream model there is, and worth it only when the hardest reasoning genuinely pays for itself.
At the cheap end the direction flips. GPT-4o mini at $0.15 in and $0.60 out is more than five times cheaper than Claude Haiku 3.5 at $0.80 and $4.00. For high-volume, simple work — tagging, sorting, short replies — the small OpenAI models are markedly cheaper.
Look down the last two columns and the pattern repeats everywhere: output costs four to five times what input costs. A long prompt is cheap. A long answer is not. Capping reply length is usually the fastest way to cut a bill in half, and it costs you nothing in quality if the answers were padded anyway.
This also means comparing models on the input price alone is misleading. Work it out on a realistic prompt with a realistic reply length — which is exactly what the calculator above is for.
The questions people ask when choosing between the two.
It depends entirely which models you compare. GPT-5 and Claude Sonnet 4.5 cost exactly the same, $3 per million input tokens and $15 per million output. Claude Opus 4 is five times dearer than GPT-5, while GPT-4o mini is over five times cheaper than Claude Haiku 3.5. There is no single answer across the two families.
Of the models listed here, GPT-4o mini at $0.15 input and $0.60 output per million tokens. It suits high-volume, straightforward work. For harder tasks the mid-tier models are usually better value than paying repeatedly for a cheap model that gets it wrong.
Generating text takes far more computation than reading it. Every output token has to be produced one after another, while input can be processed together. That is why the output rate is four to five times the input rate on nearly every model.
Roughly four characters of English, so about three quarters of a word. A thousand tokens is somewhere near 750 words. The calculator above estimates tokens for whatever you paste in, so you do not have to count.
No. These are API rates, paid per token for developers calling the models from their own software. The consumer subscriptions are separate flat monthly fees and are not related to these numbers.
Less often than they used to, but they do change, and new models arrive regularly. The figures here were checked in September 2026. Always confirm against the provider before committing to a budget.
Background reading on how this actually works.
LLM API pricing explained: what you're actually paying for How to write shorter AI prompts that get the same answer