OpenAI prices the GPT family per million tokens with an input rate, a cached-input rate, an output rate and a 50% Batch API discount. There is no charge to write to the cache. The table is the full current list from OpenAI's pricing page, followed by what three real workloads cost on each model.
gpt-5.5gpt-5.4gpt-5-nanoNo charge for cache writes. Batch API is 50% off. GPT-5.4 and later, GPT-5.5 and the GPT-6 family bill prompts over 272k input tokens at roughly 2x input and 1.5x output. Pro models have no cached-input rate. Source: developers.openai.com →
| Model | Input | Cached input | Output | Batch | Notes |
|---|---|---|---|---|---|
GPT-6 Astragpt-6-astra |
$10 | $1 | $50 | 50% off | Over 272k input tokens: $20 / $75. |
GPT-6.1 Solgpt-6.1-sol |
$2 | $0.1 | $10 | 50% off | Over 272k input tokens: $4 / $15. |
GPT-6 Solgpt-6-sol |
$2 | $0.2 | $10 | 50% off | Over 272k input tokens: $4 / $15. |
GPT-6 Lunagpt-6-luna |
$0.1 | $0.01 | $0.5 | 50% off | Over 272k input tokens: $0.20 / $0.75. |
GPT-5.6 Solgpt-5.6-sol |
$4 | $0.4 | $20 | 50% off | Over 272k input tokens: $8 / $30. |
GPT-5.6 Terragpt-5.6-terra |
$2 | $0.2 | $12 | 50% off | Over 272k input tokens: $4 / $18. |
GPT-5.6 Lunagpt-5.6-luna |
$0.2 | $0.02 | $1.2 | 50% off | Over 272k input tokens: $0.40 / $1.80. |
GPT-5.5gpt-5.5 |
$5 | $0.5 | $30 | 50% off | Over 272k input tokens: $10 / $45. |
GPT-5.5 Progpt-5.5-pro |
$30 | — | $180 | 50% off | No cached-input rate. Over 272k input tokens: $60 / $270. |
GPT-5.4gpt-5.4 |
$2.5 | $0.25 | $15 | 50% off | Over 272k input tokens: $5 / $22.50. |
GPT-5.4 Minigpt-5.4-mini |
$0.75 | $0.075 | $4.5 | 50% off | |
GPT-5.4 Nanogpt-5.4-nano |
$0.2 | $0.02 | $1.25 | 50% off | |
GPT-5.4 Progpt-5.4-pro |
$30 | — | $180 | 50% off | No cached-input rate. Over 272k input tokens: $60 / $270. |
GPT-5.2gpt-5.2 |
$1.75 | $0.175 | $14 | 50% off | |
GPT-5.2 Progpt-5.2-pro |
$21 | — | $168 | 50% off | No cached-input rate. |
GPT-5.1gpt-5.1 |
$1.25 | $0.125 | $10 | 50% off | |
GPT-5gpt-5 |
$1.25 | $0.125 | $10 | 50% off | |
GPT-5 Minigpt-5-mini |
$0.25 | $0.025 | $2 | 50% off | |
GPT-5 Nanogpt-5-nano |
$0.05 | $0.005 | $0.4 | 50% off | |
GPT-5 Progpt-5-pro |
$15 | — | $120 | 50% off | No cached-input rate. |
List prices are hard to feel per million. These are three monthly workloads, priced on the models above at the standard tier, with the batch discount applied only where the workload is asynchronous. Retries, tool-call loops and prompt growth are not in these numbers, which is why estimates and invoices disagree.
| Model | Support chatbot | Tool-using agent | Overnight classification |
|---|---|---|---|
GPT-6 Astragpt-6-astra |
$2,000 | $2,925 | $5,250 |
GPT-5.5gpt-5.5 |
$1,100 | $1,575 | $2,750 |
GPT-5.4gpt-5.4 |
$550 | $788 | $1,375 |
GPT-5.4 Minigpt-5.4-mini |
$165 | $236 | $413 |
GPT-5 Nanogpt-5-nano |
$13.00 | $18.00 | $30.00 |
Take the tool-using agent above on GPT-5.4. Its 180,000,000 input tokens a month cost $450 at the input rate. If 80% of those tokens are a repeated prefix (system prompt, tool definitions, a shared document) served from the cache at $0.25, the input line falls to $126, a 72% saving before output tokens. OpenAI does not charge for cache writes. A prefix that repeats within the cache window is simply billed at the cached-input rate on the second and later requests, which makes the saving easier to predict than on providers with a write premium.
GPT-5.4 and later, GPT-5.5 and the GPT-6 family bill prompts over 272,000 input tokens at roughly 2x input and 1.5x output. Agents that carry a long history or large tool results can cross the line on a single request.
GPT-5.5 Pro, GPT-5.4 Pro, GPT-5.2 Pro and GPT-5 Pro bill every input token at the full rate, and their output rates are 6x the standard model. They are priced for a few hard calls a day, not for a production loop.
Asynchronous jobs that can wait up to 24 hours pay 50% on every token. OpenAI also publishes priority processing at a premium for latency-sensitive calls; that tier is not in the table.
GPT-4.1, GPT-4o and the o-series remain on the API at the rates shown. A codebase that still names gpt-4o is paying $2.50 input for a model that GPT-5.4 Mini undercuts at $0.75 with better results on most tasks.
Still priced because code still names them. A legacy model is available but superseded; a retired one no longer runs on this provider's API, though several remain on cloud marketplaces at the rate shown.
| Model | Input | Cached input | Output | Notes |
|---|---|---|---|---|
GPT-4.1legacygpt-4.1 |
$2 | $0.5 | $8 | |
GPT-4.1 Minilegacygpt-4.1-mini |
$0.4 | $0.1 | $1.6 | |
GPT-4.1 Nanolegacygpt-4.1-nano |
$0.1 | $0.025 | $0.4 | |
GPT-4olegacygpt-4o |
$2.5 | $1.25 | $10 | |
GPT-4o Minilegacygpt-4o-mini |
$0.15 | $0.075 | $0.6 | |
o3legacyo3 |
$2 | $0.5 | $8 | |
o3-prolegacyo3-pro |
$20 | — | $80 | No cached-input rate. |
o3-minilegacyo3-mini |
$1.1 | $0.55 | $4.4 | |
o4-minilegacyo4-mini |
$1.1 | $0.275 | $4.4 | |
o1legacyo1 |
$15 | $7.5 | $60 |
The same three roles on each provider, input and output per million tokens. Tier for tier is the only fair comparison; a budget model on one provider against a flagship on another tells you nothing.
| Provider | Flagship | Workhorse | Budget |
|---|---|---|---|
| Anthropic | $4 / $20 Claude Opus 5.5 |
$2 / $10 Claude Sonnet 5.5 |
$0.1 / $0.5 Claude Haiku 5.5 |
| OpenAI | $5 / $30 GPT-5.5 |
$2.5 / $15 GPT-5.4 |
$0.05 / $0.4 GPT-5 Nano |
| Google Gemini | $2 / $12 Gemini 3.1 Pro Preview |
$0.75 / $3.75 Gemini 3.8 Flash |
$0.3 / $2.5 Gemini 3.5 Flash-Lite |
Current GPT models range from $0.05 input and $0.4 output per million tokens on GPT-5 Nano to $30 input and $180 output on GPT-5.4 Pro. GPT-5.4 sits at $2.5 input and $15 output. Cached input is 0.1x on most models and the Batch API is 50% off.
No. ChatGPT Plus, Pro and Team are subscriptions for the ChatGPT app, billed per person per month. The OpenAI API is billed per token to developers calling the models from their own code. This page covers the API.
On GPT-5.4, a prompt that is 80% cached input costs 72% less in input tokens than the same prompt uncached ($450 against $126 a month on the worked agent example below). There is no write charge, so the saving starts on the second request.
On GPT-5.4 and later, any request whose input exceeds 272,000 tokens is billed at the higher long-context rates for the whole request, roughly 2x input and 1.5x output. The table shows the base rate and the surcharge is in the notes column.
Reasoning tokens bill as output tokens at the model's output rate. They never appear in the visible answer, but they are counted in the usage object of each response, so they belong in any cost calculation.
These are the rates. Agent Ping applies them to every real run your agents make, by agent, customer and feature, and pages you when one of them spikes. It prices OpenAI calls from this same table.
Free plan, no card. Prices on this page verified 17 October 2026.