ChatGPT API Pricing Explained (2026)

ChatGPT API pricing is based on tokens — the chunks of text your requests send and receive. As of October 2026, the main models are GPT-6 Luna ($0.10 input / $0.50 output per million tokens), GPT-6.1 Sol ($2.00 / $10.00), and GPT-6 Astra ($10.00 / $50.00). You only pay for what you use, with no monthly minimum.

What Is ChatGPT API Pricing and How Does It Work?

ChatGPT API pricing is charged per token. A token is roughly four characters of text, so a short sentence might be 15 to 20 tokens. Every API call has an input token count (the text you send, including any instructions and conversation history) and an output token count (the text the model generates back).

You are billed separately for input and output. Output tokens cost more than input tokens on most models, because generating text requires more compute than reading it. You only pay for what you use, and there is no subscription or monthly minimum on the standard API.

What You Will Learn in This Guide

• How token-based pricing works

• The current model tiers and their prices

• Which model to choose for your use case

• How to estimate your costs before you build

• How to control your spending with usage limits

• The Batch API, Flex, and Fast mode explained

How Token Pricing Works

Every piece of text you send to the API, and every response you receive, is measured in tokens before billing is applied. OpenAI's tokeniser splits text at word boundaries, punctuation, and common subwords. As a rough guide:

• 1 token is roughly 4 characters or 0.75 words in English

• 100 tokens is about 75 words, or a short paragraph

• 1,000 tokens is roughly three-quarters of a page of text

• 1 million tokens is a large novel

Prices are listed per million tokens. To work out the cost of a typical request, estimate your input size (your prompt plus any conversation history) and your expected output size (the model's reply), then multiply each by the relevant rate.

Input vs Output Tokens

Input tokens are the text you send to the model. This includes your system prompt, the user's message, any context you have added, and any conversation history you are passing in. Output tokens are the tokens the model generates in response.

Output tokens cost more than input tokens. On GPT-6 Luna, for example, input is $0.10 per million tokens and output is $0.50 per million, a 5:1 ratio. This ratio is consistent across all three main model tiers.

Cached Input Tokens

If you send the same prompt prefix repeatedly, such as a long system prompt that stays the same across every request, OpenAI caches it automatically. Cached input tokens are charged at a much lower rate: 90% less on GPT-6 Luna ($0.01 per million), GPT-6.1 Sol ($0.10 per million), and GPT-6 Astra ($1.00 per million). This can significantly reduce costs for applications that reuse the same instructions across thousands of calls.

The Current ChatGPT API Models and Prices (October 2026)

There are three main model tiers available on the standard API, designed for different use cases and budgets.

GPT-6 Luna: The Efficient Model

Best for: high-volume, straightforward tasks where speed and low cost matter most.

• Input: $0.10 per million tokens

• Cached input: $0.01 per million tokens

• Output: $0.50 per million tokens

GPT-6 Luna is OpenAI's most efficient model. It is fast and inexpensive, making it the right choice for classification, summarisation, simple Q&A, and content moderation at scale. If you are building a feature that handles thousands of short requests per day, Luna is usually the starting point.

GPT-6.1 Sol: The Balanced Model

Best for: coding, complex reasoning, and agentic workflows where capability matters more than raw cost.

• Input: $2.00 per million tokens

• Cached input: $0.10 per million tokens

• Output: $10.00 per million tokens

GPT-6.1 Sol sits in the middle of the range. It handles complex coding tasks, multi-step reasoning, and agentic workflows well. At 20x the price of Luna, it is used where higher capability genuinely improves the output. OpenAI recommends it as the primary model for developer-focused applications.

GPT-6 Astra: The Frontier Model

Best for: the hardest end-to-end tasks, deep research, and enterprise workflows where accuracy is critical.

• Input: $10.00 per million tokens

• Cached input: $1.00 per million tokens

• Output: $50.00 per million tokens

GPT-6 Astra is OpenAI's most capable model, built for complex, multi-step tasks that require the highest level of reasoning and instruction-following. At $50 per million output tokens it is the most expensive option. Most developers use it selectively, for the hardest tasks in a pipeline, rather than for every call.

Specialised Models

Alongside the main three, OpenAI also offers specialised models for specific workloads:

• gpt-5.3-codex: Optimised for coding tasks. $1.75 input / $14.00 output per million tokens (standard). Powers Codex in ChatGPT Work.

• chat-latest: The model behind ChatGPT conversations, accessible via API. $5.00 input / $30.00 output per million tokens.

• gpt-5.6-sol: Cyber and security model at $4.00 input / $20.00 output. Promotional pricing confirmed through at least November 2026.

• gpt-rosalind-research and gpt-rosalind-discovery: Life sciences models at $5.00 input / $25.00 output, available to approved research organisations.

Processing Modes: Standard, Batch, Flex, Fast, and Ultrafast

OpenAI offers five processing modes that trade speed for cost, or cost for speed. Choosing the right mode can make a substantial difference to your monthly bill.

Standard Mode

Standard mode is the default. You pay the listed price and receive responses at normal speed with high availability. This is suitable for most production applications where response time matters.

Batch API (50% Discount)

The Batch API processes requests asynchronously, with results delivered within 24 hours. In exchange, you pay 50% less on both input and output tokens. This is ideal for bulk processing jobs where you do not need an immediate reply: generating summaries for a large document library, running content moderation across thousands of items, or analysing datasets overnight.

• GPT-6 Luna Batch: $0.05 input / $0.25 output per million tokens

• GPT-6.1 Sol Batch: $1.00 input / $5.00 output per million tokens

• GPT-6 Astra Batch: $5.00 input / $25.00 output per million tokens

Flex Processing (50% Discount)

Flex processing also offers a 50% discount with the same pricing as Batch, but is designed for lower-priority tasks that can tolerate occasional resource unavailability and slower response times. It is useful for non-production workloads, internal tools, and development testing.

Fast Mode (2x Price)

Fast mode guarantees high-speed, reliable performance and charges a 2x premium over standard pricing. It was previously called Priority processing and was renamed Fast mode on 30 July 2026. Use it when your application requires consistently low latency, such as real-time interactive tools where a slow response would frustrate users.

Ultrafast Mode (6x Price)

Ultrafast mode charges 6x the standard rate in exchange for the lowest possible response times. Most applications will not need this tier, but it exists for high-frequency applications, real-time voice experiences, and enterprise systems where every millisecond matters.

How to Estimate Your API Costs Before You Build

Before committing to a model, it is worth estimating your monthly costs based on expected usage. Here is a straightforward method.

Step 1: Estimate Your Token Usage Per Request

Count the typical length of your system prompt, the average user message length, any context you pass in, and the average length of the response you expect. Add these together to get a rough token count per request.

For example, a customer support bot might have a 500-token system prompt, a 100-token user message, and generate a 200-token reply. That is 600 input tokens and 200 output tokens per request.

Step 2: Multiply by Your Expected Volume

Multiply your per-request token counts by the number of requests you expect per month. If your support bot handles 10,000 conversations per month:

• Input: 600 tokens x 10,000 = 6 million input tokens

• Output: 200 tokens x 10,000 = 2 million output tokens

Step 3: Apply the Model Price

Using GPT-6 Luna at $0.10 input and $0.50 output per million tokens:

• Input cost: 6M tokens x $0.10 = $0.60

• Output cost: 2M tokens x $0.50 = $1.00

• Total: $1.60 per month

The same volume on GPT-6.1 Sol would cost $12 input + $20 output = $32 per month. On GPT-6 Astra it would be $60 + $100 = $160 per month. Model selection is the single most impactful cost decision you can make.

Which ChatGPT API Model Should You Use?

The right model depends entirely on the task. Here is a plain-English guide:

• Classification, tagging, and simple Q&A: GPT-6 Luna. Fast and cheap. The accuracy gap over Sol is small for structured tasks.

• Code generation and debugging: GPT-6.1 Sol. Strong coding performance at a reasonable price for developer tools.

• Complex reasoning and multi-step agents: GPT-6 Astra or GPT-6.1 Sol. Astra for the hardest tasks; Sol for most agentic workflows.

• Bulk overnight processing: Any model via the Batch API at 50% off.

• Real-time voice conversations: GPT-Live-1 at $0.05 per minute.

• Coding workflows (Codex-style): gpt-5.3-codex at $1.75 input / $14.00 output.

OpenAI recommends testing models in the Playground before committing to one for production. Playground usage is billed at the same per-token rate as API calls.

If you are building your first API project, start with GPT-6 Luna. It is inexpensive enough that you can experiment without worrying about cost, and you can move up to Sol or Astra once you know exactly where the capability gaps are.

How to Control Your API Spending

The OpenAI platform includes built-in tools for keeping costs under control.

Monthly Spend Limits

You can set a monthly budget in your billing settings at platform.openai.com. Once your spend reaches the limit, OpenAI stops serving requests. There may be a short delay in enforcing the cap, and you are responsible for any small overage. Set the limit slightly above your expected usage, rather than exactly at it, to avoid unintended outages.

Email Alerts

You can configure a notification threshold separately from the hard limit. When your spend crosses the threshold, OpenAI sends an email alert. This is useful for catching unexpected cost spikes early, before they hit your hard cap.

Usage Dashboard

The usage tracking dashboard at platform.openai.com shows your token consumption and spend for the current and past billing cycles, broken down by model. It is the most reliable way to spot which models or projects are driving the highest costs.

Project-Level Limits

If you use OpenAI Projects to organise your API work, you can set billing restrictions per project. This lets teams working on different products each have their own spending cap, separate from the organisation total.

Other API Costs to Be Aware Of

Beyond the main text models, several tools carry their own pricing on top of the per-token rate:

• Web search: $10.00 per 1,000 calls. Search content tokens are billed at the model's standard rates on top of the call fee.

• Code Containers (Hosted Shell / Code Interpreter): $0.03 for a 1 GB container per 20-minute session, up to $1.92 for 64 GB.

• File search (Responses API): $0.10 per GB per day for storage, plus $2.50 per 1,000 tool calls.

• Image generation (GPT-Image-2.5): $8.00 per million image input tokens, $30.00 per million output tokens.

• Data residency: A 10% pricing uplift applies to models released on or after 5 March 2026.

Frequently Asked Questions About ChatGPT API Pricing

Does ChatGPT Plus include API access?

No. ChatGPT Plus, Business, Enterprise, and Edu subscriptions are entirely separate from API access. The API is billed on a pay-as-you-go basis through your OpenAI developer account at platform.openai.com. A ChatGPT subscription does not give you API credits or reduce API costs.

What is a token, and how many tokens are in a typical message?

A token is roughly four characters of English text, or about three-quarters of a word. A short conversational message of 50 words is approximately 65 to 70 tokens. A 500-word article is around 650 tokens. The exact count depends on the words used: common words tokenise more efficiently than rare or technical terms. OpenAI provides a tokeniser tool in the Playground where you can count tokens precisely before building.

Is the Batch API suitable for production applications?

It depends on the use case. The Batch API processes requests within 24 hours, which makes it unsuitable for anything requiring a real-time response. For offline processing, bulk content generation, nightly analysis runs, or large data pipelines, it is well-suited and saves 50% on costs compared to standard pricing.

How do I avoid unexpected API costs?

Set a monthly spend limit in your billing settings and add an email alert threshold slightly below it. Monitor the usage dashboard regularly. Use the cheapest model that meets your quality requirements, and consider the Batch API for non-urgent workloads. Prompt caching also reduces costs significantly if your system prompts are long and consistent across requests.

Does OpenAI charge for failed API requests?

You are charged for tokens your request consumed up to the point it was processed. If a request fails before the model begins processing, you should not be charged. If it fails partway through generation, you may be charged for tokens generated before the failure. Check the usage dashboard after any incident to confirm the exact charge.

Can I get a volume discount on the API?

The Batch API gives a flat 50% discount at all volume levels with no minimum. For very large workloads, OpenAI offers Reserved Capacity and Scale Tier through their enterprise sales team, designed for organisations with sustained, high-volume usage. You can also connect OpenAI models through Amazon Bedrock and Microsoft Azure, which are billed through those platforms.

The Bottom Line on ChatGPT API Pricing

ChatGPT API pricing is pay-as-you-go and token-based. The three main models, GPT-6 Luna, GPT-6.1 Sol, and GPT-6 Astra, cover most use cases from simple classification to complex enterprise reasoning. Choosing the right model is the single most important cost decision you can make.

For most developers starting out, GPT-6 Luna for high-volume tasks and GPT-6.1 Sol for anything needing genuine reasoning is a solid combination. Use the Batch API whenever results within 24 hours are acceptable, and set spend limits from day one to avoid surprises.

As OpenAI continues to release new models, pricing will evolve. Check the official pricing page before making model decisions for new projects.

Explore more ChatGPT guides and developer resources on whatischatgpt.co.uk.

References

1. OpenAI API Pricing — official pricing page for all models and tools.

2. OpenAI API Documentation: Pricing — detailed pricing tables including processing modes and specialised models.

3. OpenAI Platform: Batch API Guide — how the Batch API works and how to use it to reduce costs.