Why pricing has two numbers

Every model on ModelDex has two prices, not one: a price per million input tokens and a price per million output tokens. Input is everything you send the model. Output is everything it generates. They are billed at different rates, and across every provider we track, output is the more expensive of the two, often by a wide margin. That single fact drives most of the cost decisions you will make.

Prices are quoted per million tokens, written as per Mtok. A token is roughly three quarters of an English word. So a thousand word prompt is very roughly thirteen hundred tokens, and a model priced at $5 per million input tokens costs about two thirds of a cent to read that prompt. The numbers feel small per call and large at volume. Volume is where pricing decisions are won or lost.

The spread is enormous

The gap between the cheapest and most expensive models we track is not small. It is two orders of magnitude.

On input, GPT-5.4 nano is $0.20 per million tokens and Gemini 2.5 Flash is $0.30. At the top, GPT-5.5 Pro is $30.00 per million input tokens. That is a hundred and fifty fold spread on input alone.

On output, the spread is wider still. GPT-5.4 nano is $1.25 per million output tokens. GPT-5.5 Pro is $180.00. That is a roughly hundred and forty fold difference for the same unit of generated text. Choosing the wrong tier for a high volume task is the most common way teams overspend on models.

A tour of the price ladder

At the bottom sit the small, fast models. GPT-5.4 nano is $0.20 input and $1.25 output. Gemini 2.5 Flash is $0.30 and $2.50. Gemini 3 Flash is $0.50 and $3.00. GPT-5.4 mini is $0.75 and $4.50. Claude Haiku 4.5 is $1 and $5. Grok Build 0.1 is $1.00 and $2.00. These are the models you run at scale.

In the middle sit the workhorses. Grok 4.3 is unusually cheap for its class at $1.25 input and $2.50 output. Gemini 3.1 Pro is $2.00 and $12.00 for prompts at or under 200,000 tokens. GPT-5.4 is $2.50 and $15.00. Claude Sonnet 4.6 is $3 and $15. Gemini 3.5 Flash is $1.50 and $9.00.

At the top sit the flagships. Claude Opus 4.8 is $5 input and $25 output. GPT-5.5 is $5 and $30. And then GPT-5.5 Pro stands apart at $30 input and $180 output, in a class of its own on price.

Output dominates your bill

For most generative work, the model writes more than you might expect, especially reasoning models that produce long internal chains of thought before the final answer. Because output is billed several times higher than input on every provider here, your bill is usually driven by output volume, not input volume.

Compare two flagships. Claude Opus 4.8 at $25 output and GPT-5.5 at $30 output are close. But GPT-5.5 Pro at $180 output is six times either of them. If a task generates long answers at scale, that output multiple is the whole story. Estimate output tokens carefully, because that is the number that moves your invoice.

How to estimate a real cost

Write down four numbers. The average input tokens per call. The average output tokens per call. The model's input and output prices. The number of calls per month.

Cost per call is input tokens divided by one million, times the input price, plus output tokens divided by one million, times the output price. Multiply by monthly calls for the monthly total. Do this for two or three candidate models before you commit. The arithmetic frequently shows that a mid-tier model at a fraction of the flagship output price clears the task at a fraction of the cost.

Things that change the headline price

A few realities sit underneath the sticker numbers. Long prompts can be billed differently: Gemini 2.5 Pro and Gemini 3.1 Pro quote their input prices for prompts at or under 200,000 tokens, which signals a separate rate for longer prompts. Many providers also offer prompt caching and batch discounts that lower the effective rate for repeated or non urgent work. And reasoning effort raises output token counts, which raises cost even though the per token price is unchanged. Check the provider pricing page for the specific feature you plan to use.

Where the numbers come from

Every price in this guide is a verified figure on the matching ModelDex model page, traced to the provider's own pricing documentation: Anthropic, OpenAI, Google, and xAI. Providers change prices, and our dataset tracks their pricing pages, so the live model page is always the current rate.