How Much Does ChatGPT API Cost? | Current Price Breakdown

If you want one clean answer, the ChatGPT API can be cheap or pricey depending on the model, the size of your prompt, and how long the reply runs. A lightweight app can cost pennies across many calls. A high-end reasoning workflow can climb fast once output tokens stack up.

That’s why “How much does ChatGPT API cost?” doesn’t have one flat number. OpenAI charges by usage, not by a single monthly API fee for most self-serve accounts. You pay for input tokens, cached input when it applies, output tokens, and, in some cases, extra tool charges.

The part that catches many people off guard is this: output usually costs more than input, and premium models can jump far beyond the low-cost tiers. So the real question is not just the posted rate. It’s what your app sends, what it asks back, and how often it calls the API.

How Much Does ChatGPT API Cost? By Model Tier

As of March 2026, OpenAI’s flagship text pricing starts at $0.20 per 1 million input tokens for GPT-5.4 nano, moves to $0.75 for GPT-5.4 mini, and reaches $2.50 for GPT-5.4. Output pricing is steeper: $1.25, $4.50, and $15.00 per 1 million output tokens for those same tiers.

At the top end, GPT-5.4 Pro is priced for heavier work. Its standard rate is $30.00 per 1 million input tokens and $180.00 per 1 million output tokens. That tier makes sense only when the added reasoning quality pays for itself in your product or workflow.

What You’re Paying For

OpenAI token billing usually breaks into three parts:

• Input tokens: the text, instructions, and context you send in.
• Cached input tokens: repeated prompt content that can be billed at a lower rate.
• Output tokens: the model’s reply, which often costs more than the prompt.

That split matters. A short user message with a long generated reply can cost more than people expect. A giant system prompt reused across many calls may cost less than it looks if caching kicks in. OpenAI’s API Pricing page lists the current model rates, while the developer pricing docs break out standard, batch, flex, and priority processing tiers.

What A Small Request Looks Like In Dollars

Say your app sends 1,000 input tokens and gets back 500 output tokens on GPT-5.4 mini. At posted standard rates, that works out to a tiny fraction of a cent per call. Run that thousands of times per day, though, and the monthly total starts to matter.

Now switch that same traffic to GPT-5.4 Pro and the math changes in a hurry. The result may still be fine for a legal drafting flow, a coding reviewer, or a research-heavy agent. It’s overkill for a simple FAQ bot or a short classification task.

Current Model Prices At A Glance

The table below gives a practical side-by-side view of the main API rates many builders care about first.

Model Or Service	Input Price	Output Price
GPT-5.4 nano	$0.20 / 1M tokens	$1.25 / 1M tokens
GPT-5.4 mini	$0.75 / 1M tokens	$4.50 / 1M tokens
GPT-5.4	$2.50 / 1M tokens	$15.00 / 1M tokens
GPT-5.4 Pro	$30.00 / 1M tokens	$180.00 / 1M tokens
GPT-realtime-1.5 text	$4.00 / 1M tokens	$16.00 / 1M tokens
GPT-realtime-mini text	$0.60 / 1M tokens	$2.40 / 1M tokens
Web search tool	$10.00 / 1K calls	Tool fee only
File search tool	$2.50 / 1K calls	Plus storage fees

Those posted numbers are only the start. OpenAI also lists lower-cost batch processing and flex processing for some models, plus priority processing for apps that need faster and steadier throughput. The detailed pricing tables are worth checking before you lock in a model for production.

Where Costs Climb Faster Than People Expect

Most surprise bills come from one of four places.

Long Replies

Output tokens can be the sneaky part of the bill. If your app asks for full essays, long JSON blobs, or multi-step reasoning traces, the reply can cost more than the prompt that triggered it.

Big Prompt Scaffolding

Many apps ship with a huge system prompt, examples, policy text, and retrieved context on every request. That can work well, yet it pushes token counts up fast. Tightening those inputs often saves money with no drop in quality.

Premium Models Used For Routine Work

A lot of tasks do not need the top model. Labeling support tickets, cleaning product data, extracting fields, and writing short summaries often fit better on mini or nano tiers. Save the priciest model for work where accuracy gains pay for the extra spend.

Extra Tool Charges

If your app uses web search, file search, containers, realtime audio, or image generation, your invoice is no longer just a plain token bill. Those extras can still be worth it. You just want to price them on purpose, not by accident.

• Trim prompt boilerplate that repeats in every call.
• Set sane output limits so replies don’t ramble.
• Route easy jobs to mini or nano models.
• Use caching when the same context shows up again and again.
• Test batch or flex pricing for jobs that don’t need instant replies.

What A Monthly Budget Might Look Like

Monthly spend depends on traffic volume more than anything else. A solo tool used a few hundred times per day may stay modest. A customer-facing app with thousands of sessions, long chat histories, and tool calls can rack up a much larger bill.

A simple way to sketch a budget is to multiply your average input and output tokens per request by your expected request count, then match that against the posted price for the model you plan to run. Do that with two or three model choices, not just one. The gap can be wide.

Usage Pattern	Likely Cost Shape	What Usually Helps
Short prompts, short replies, high volume	Nano or mini can stay lean	Keep prompts tight and reuse cached context
Medium prompts, polished long replies	Output cost starts to lead	Shorten reply targets and trim examples
Heavy reasoning or premium coding flows	Top-tier costs rise fast	Reserve Pro for the narrow slice that needs it
Agents with web or file tools	Token bill plus tool fees	Track tool calls as a separate budget line
Large back-office jobs	Standard rates may be wasteful	Shift eligible jobs to batch or flex

How To Estimate Your Own Spend Without Guessing

If you want a cleaner estimate, use a short process:

1. Pick one real task from your app, not a made-up sample.
2. Measure the average input tokens that task sends.
3. Measure the average output tokens it gets back.
4. Run the math against two or three model tiers.
5. Add any tool fees your app triggers.
6. Multiply by daily and monthly traffic.

Do this with your actual prompt format and actual reply length. That’s the part that gives you a useful budget, since token counts from toy prompts rarely match production traffic.

Don’t Mix ChatGPT Plan Pricing With API Billing

This trips up a lot of buyers. ChatGPT subscriptions and API billing are handled as separate products. Paying for ChatGPT Plus, Pro, Business, or Enterprise does not mean your API usage is folded into the same billing flow. OpenAI’s billing settings in ChatGPT vs Platform note spells that out plainly.

So if you’re pricing a product, a client workflow, or an internal app, use the API rate sheet and your usage data. Don’t rely on what you pay for the ChatGPT app each month.

Which Model Usually Fits Which Job

GPT-5.4 nano makes sense when volume is high and each task is simple. Think classification, tagging, short rewrites, or light extraction. GPT-5.4 mini is often the sweet spot for many apps because it stays far cheaper than the flagship tier while still handling a broad set of real business tasks well.

GPT-5.4 is the step up when answer quality, reasoning depth, or tool-heavy work matters more. GPT-5.4 Pro belongs in a narrow lane: hard work where weaker models create enough rework, misses, or manual cleanup that the higher price still comes out ahead.

That’s the cleanest way to think about ChatGPT API cost: the cheapest model is not always the cheapest finished workflow, and the strongest model is not always the smartest buy.