OpenAI API pricing: OpenAI price page, September 2026
gpt-5-nano has the lowest listed input price at $0.05 per million tokens; o1-pro has the highest at $150. Rates come from OpenAI's expanded pricing tables, checked 28 September 2026.
OpenAI’s cheapest API language model is gpt-5-nano, at $0.05 per million input tokens, $0.005 for cached input and $0.40 for output, on OpenAI’s pricing page checked 28 September 2026.1 The most expensive is o1-pro, at $150.00 input and $600.00 output.1 The current GPT-6 line sits between them: Luna at $0.10, Sol at $2.00 and Astra at $10.00 per million input tokens.2
Output tokens and cache hits change the bill alongside the input rate.
TL;DR
- Output drives text bills. On every GPT-5 and GPT-6 row, output costs five to eight times the input rate.1
- Cached input costs a tenth of the input rate on every GPT-5 and GPT-6 model that lists one. On gpt-4.1 and o3 it’s a quarter, and on gpt-4o and o1 it’s half. Pro models list no cached rate.1
- Batch usually halves input and output; gpt-3.5-turbo-1106 carries no discount. Flex matches Batch input and output rates. Fast mode costs 1.67 to 2.5 times Standard.1
- Long prompts cost more. On the ten models with long-context rates, input doubles and output rises by half.1
- On price alone, GPT-6 Sol costs half of GPT-5.6 Sol’s promotional rate, and GPT-6 Luna costs half or less of GPT-5.6 Luna.1
- Audio, image and embedding models use other units: realtime audio from $10.00 per million input tokens, transcription from $0.003 a minute and embeddings from $0.02 per million tokens.1
How this ranking was built
This page lists every OpenAI API language model’s input, cached input and output price per million tokens from OpenAI’s pricing pages, checked 28 September 2026, ordered by input price. The ranked field is OpenAI’s Input column: Standard processing, short context, per 1M tokens, from the API docs pricing page captured with every “All models” list expanded.1 It belongs to our model infrastructure coverage.
- Take every row with a per-token input price for a text model from four tables: Flagship models (10 rows), its expanded All models list (29 rows), Cyber models (3 rows) and Specialized models (8 rows).1
- Merge gpt-5.6-sol, which appears in both the Flagship and Cyber tables at identical rates. The Cyber table adds 2 rows, for 41.1
- Move the three embedding models and omni-moderation-latest to their own section. The Specialized table adds 4 rows, for 45.
- Sort by input price, lowest first. Ties go to the lower output price, then to the model ID.
Two other OpenAI captures from the same day, openai.com/api/pricing and platform.openai.com/docs/pricing, show the same GPT-6 rates.2,3 OpenAI prints every rate with a dollar sign and states no currency code or tax treatment on these pages.1 These are OpenAI’s listed prices: company statements, not measured bills. Our research standard explains how Glamdring treats company figures.
What does each OpenAI model cost per million tokens?
OpenAI lists 45 API language models, with input prices from $0.05 per million tokens for gpt-5-nano to $150.00 for o1-pro, on its pricing page checked 28 September 2026. Output runs from $0.40 to $600.00. The table gives Standard short-context rates and OpenAI’s Batch rates per 1M tokens. A dash means OpenAI lists no rate.1
| Rank | Model | OpenAI section | Input ($/1M) | Cached input ($/1M) | Output ($/1M) | Batch input ($/1M) | Batch output ($/1M) |
|---|---|---|---|---|---|---|---|
| 1 | gpt-5-nano | All models | $0.05 | $0.005 | $0.40 | $0.025 | $0.20 |
| 2 | gpt-4.1-nano | All models | $0.10 | $0.025 | $0.40 | $0.05 | $0.20 |
| 3 | gpt-6-luna | Flagship | $0.10 | $0.01 | $0.50 | $0.05 | $0.25 |
| 4 | gpt-4o-mini | All models | $0.15 | $0.075 | $0.60 | $0.075 | $0.30 |
| 5 | gpt-5.6-luna | Flagship | $0.20 | $0.02 | $1.20 | $0.10 | $0.60 |
| 6 | gpt-5.4-nano | All models | $0.20 | $0.02 | $1.25 | $0.10 | $0.625 |
| 7 | gpt-5-mini | All models | $0.25 | $0.025 | $2.00 | $0.125 | $1.00 |
| 8 | babbage-002 | All models | $0.40 | - | $0.40 | $0.20 | $0.20 |
| 9 | gpt-4.1-mini | All models | $0.40 | $0.10 | $1.60 | $0.20 | $0.80 |
| 10 | gpt-3.5-turbo | All models | $0.50 | - | $1.50 | - | - |
| 11 | gpt-3.5-turbo-0125 | All models | $0.50 | - | $1.50 | $0.25 | $0.75 |
| 12 | gpt-5.4-mini | All models | $0.75 | $0.075 | $4.50 | $0.375 | $2.25 |
| 13 | gpt-3.5-turbo-1106 | All models | $1.00 | - | $2.00 | $1.00 | $2.00 |
| 14 | o3-mini | All models | $1.10 | $0.55 | $4.40 | $0.55 | $2.20 |
| 15 | o4-mini | All models | $1.10 | $0.275 | $4.40 | $0.55 | $2.20 |
| 16 | gpt-5 | All models | $1.25 | $0.125 | $10.00 | $0.625 | $5.00 |
| 17 | gpt-5-search-api | Specialized | $1.25 | $0.125 | $10.00 | - | - |
| 18 | gpt-5.1 | All models | $1.25 | $0.125 | $10.00 | $0.625 | $5.00 |
| 19 | gpt-3.5-turbo-instruct | All models | $1.50 | - | $2.00 | - | - |
| 20 | gpt-5.2 | All models | $1.75 | $0.175 | $14.00 | $0.875 | $7.00 |
| 21 | gpt-5.3-codex | Specialized | $1.75 | $0.175 | $14.00 | - | - |
| 22 | davinci-002 | All models | $2.00 | - | $2.00 | $1.00 | $1.00 |
| 23 | gpt-4.1 | All models | $2.00 | $0.50 | $8.00 | $1.00 | $4.00 |
| 24 | o3 | All models | $2.00 | $0.50 | $8.00 | $1.00 | $4.00 |
| 25 | gpt-6-sol | Flagship | $2.00 | $0.20 | $10.00 | $1.00 | $5.00 |
| 26 | gpt-5.6-terra | Flagship | $2.00 | $0.20 | $12.00 | $1.00 | $6.00 |
| 27 | gpt-4o | All models | $2.50 | $1.25 | $10.00 | $1.25 | $5.00 |
| 28 | gpt-5.4 | Flagship | $2.50 | $0.25 | $15.00 | $1.25 | $7.50 |
| 29 | gpt-5.6-sol | Flagship | $4.00 | $0.40 | $20.00 | $2.00 | $10.00 |
| 30 | gpt-4o-2024-05-13 | All models | $5.00 | - | $15.00 | $2.50 | $7.50 |
| 31 | gpt-rosalind-research | Specialized | $5.00 | $0.50 | $25.00 | - | - |
| 32 | chat-latest | Specialized | $5.00 | $0.50 | $30.00 | - | - |
| 33 | gpt-5.5 | Flagship | $5.00 | $0.50 | $30.00 | $2.50 | $15.00 |
| 34 | gpt-4-turbo-2024-04-09 | All models | $10.00 | - | $30.00 | $5.00 | $15.00 |
| 35 | gpt-6-astra | Flagship | $10.00 | $1.00 | $50.00 | $5.00 | $25.00 |
| 36 | gpt-5.5-cyber | Cyber | $12.50 | $1.25 | $75.00 | - | - |
| 37 | gpt-5.6-cyber | Cyber | $12.50 | $1.25 | $75.00 | - | - |
| 38 | o1 | All models | $15.00 | $7.50 | $60.00 | $7.50 | $30.00 |
| 39 | gpt-5-pro | All models | $15.00 | - | $120.00 | $7.50 | $60.00 |
| 40 | o3-pro | All models | $20.00 | - | $80.00 | $10.00 | $40.00 |
| 41 | gpt-5.2-pro | All models | $21.00 | - | $168.00 | $10.50 | $84.00 |
| 42 | gpt-4-0613 | All models | $30.00 | - | $60.00 | $15.00 | $30.00 |
| 43 | gpt-5.4-pro | Flagship | $30.00 | - | $180.00 | $15.00 | $90.00 |
| 44 | gpt-5.5-pro | Flagship | $30.00 | - | $180.00 | $15.00 | $90.00 |
| 45 | o1-pro | All models | $150.00 | - | $600.00 | $75.00 | $300.00 |
Four rows carry conditions. gpt-5.6-sol’s $4.00 and $20.00 are promotional, and OpenAI says that pricing is available at least through November 21, 2026.1 Billing for gpt-rosalind-research begins on October 5, 2026, and access is limited to approved research through OpenAI’s trusted-access program.1 The two cyber models belong to OpenAI’s Daybreak program, whose aliases gpt-daybreak-blue-latest and gpt-daybreak-red-latest currently point to gpt-5.6-sol and gpt-5.6-cyber and take the underlying model’s price.1
The input order hides how far output spreads. gpt-5-nano, gpt-4.1-nano and babbage-002 all charge $0.40 per million output tokens, while gpt-6-luna, at rank 3, charges $0.50.1 Further up, gpt-6-sol and gpt-4.1 share a $2.00 input rate, but gpt-6-sol’s output costs $10.00 against gpt-4.1’s $8.00.1 A workload that writes more than it reads should be sorted by the output column, not the rank.
What do OpenAI’s latest models cost, and when do long prompts cost more?
GPT-6 Sol and Luna have lower list prices than their GPT-5.6 counterparts. GPT-6 Sol costs $2.00 input and $10.00 output per million tokens, half of GPT-5.6 Sol’s promotional $4.00 and $20.00. GPT-6 Luna’s $0.10 and $0.50 is half of GPT-5.6 Luna’s input and less than half of its output. GPT-6 Astra, at $10.00 and $50.00, costs twice GPT-5.5’s input. These are short-context rates. For long-context requests, all ten models in OpenAI’s flagship table double their input price and raise output by half.1
OpenAI describes Astra as “built for the hardest end-to-end work”, Sol as built for “complex coding and agentic workflows” and Luna as its “most efficient model for focused, high-volume tasks”.2 Those are OpenAI’s descriptions. This page compares price, not capability.
OpenAI’s product page says its GPT-6 prices reflect standard processing for context lengths under 272K.2 The docs page prints the long-context rates beside the short-context ones:
| Model | Cache writes, short ($/1M) | Long-context input ($/1M) | Long-context cached input ($/1M) | Long-context output ($/1M) |
|---|---|---|---|---|
| gpt-6-astra | $12.50 | $20.00 | $2.00 | $75.00 |
| gpt-6-sol | $2.50 | $4.00 | $0.40 | $15.00 |
| gpt-6-luna | $0.125 | $0.20 | $0.02 | $0.75 |
| gpt-5.6-sol | $5.00 | $8.00 | $0.80 | $30.00 |
| gpt-5.6-terra | $2.50 | $4.00 | $0.40 | $18.00 |
| gpt-5.6-luna | $0.25 | $0.40 | $0.04 | $1.80 |
| gpt-5.5 | - | $10.00 | $1.00 | $45.00 |
| gpt-5.5-pro | - | $60.00 | - | $270.00 |
| gpt-5.4 | - | $5.00 | $0.50 | $22.50 |
| gpt-5.4-pro | - | $60.00 | - | $270.00 |
GPT-6 and GPT-5.6 charge for cache writes at 1.25 times the input rate. GPT-5.5 and GPT-5.4 list no cache-write price.1 A cached read on GPT-6 and GPT-5.6 costs a tenth of input.1 The estimate assumes the write replaces the first request’s normal input charge. On that basis, two requests that share a prefix cost 1.35 times the input rate instead of 2.0, and each further reuse adds 0.1. Confirm the billing in OpenAI’s caching documentation before relying on it.
Three conditions sit outside the tables. Data residency endpoints add 10% for eligible models released on or after March 5, 2026. FedRAMP endpoints add 10% over the standard rates.1 OpenAI models bought through Amazon Bedrock are billed by AWS, and OpenAI says Bedrock pricing in commercial regions matches its direct pricing.1
How much do cached input and the Batch API save?
Cached input costs a tenth of the input rate on every GPT-5 and GPT-6 model that lists one. The Batch API usually halves input and output for jobs that run asynchronously over 24 hours; gpt-3.5-turbo-1106 is a listed exception.1,2 Older models get smaller cache discounts. Cached input is 25% of input on gpt-4.1, gpt-4.1-mini, gpt-4.1-nano, o3 and o4-mini, and 50% on gpt-4o, gpt-4o-mini, o1 and o3-mini. The pro models list no cached rate.1
A worked example shows the scale. One million input tokens on gpt-6-sol cost $2.00. If 800,000 of them hit the cache, the bill becomes $0.40 for the 200,000 fresh tokens plus $0.16 for the cached ones: $0.56, or 72% less.1 The estimate assumes the cached prefix was already written. Batch then halves the remaining rates: gpt-6-sol’s Batch cached input is $0.10 and its Batch output is $5.00.1 Our explainer on what controls AI inference cost covers the wider drivers behind these rates.
OpenAI offers three other processing options on the same models:
- Flex trades slower responses and occasional unavailability for lower prices.2 For every model in OpenAI’s Flex table, the input and output rates equal the Batch rates.1
- Fast mode is the tier OpenAI called Priority processing until it renamed it on July 30, 2026.1 It costs twice Standard on GPT-6 and GPT-5.6. Across the models it lists, the premium runs from 1.67 times on gpt-4o-mini ($0.25 input, $1.00 output) to 2.5 times on gpt-5.5 ($12.50 input, $75.00 output).1
- Data residency adds 10%, and for GPT-6 Sol and Luna, EU data residency is available only with Standard processing.1
Batch has gaps. OpenAI’s Batch table omits gpt-3.5-turbo, gpt-3.5-turbo-instruct, the cyber models and the specialised models, and it lists no cached rate for gpt-4.1 through o1-pro.1 One listed row carries no discount: gpt-3.5-turbo-1106 costs $1.00 and $2.00 in both Standard and Batch.1
What do OpenAI’s audio, realtime and transcription models cost?
OpenAI’s full-size realtime voice models charge $32.00 per million audio input tokens and $64.00 per million audio output tokens, and the mini versions charge $10.00 and $20.00. Transcription and live translation run from $0.003 to $0.034 a minute. GPT-Live 1 voice sessions cost $0.05 a minute, billed per second without rounding up, with backend model and tool usage charged separately.1
Realtime and audio generation, per 1M tokens:1
| Model | Audio input | Audio cached input | Audio output | Text input | Text cached input | Text output |
|---|---|---|---|---|---|---|
| gpt-realtime-2.1 | $32.00 | $0.40 | $64.00 | $4.00 | $0.40 | $24.00 |
| gpt-realtime-2 | $32.00 | $0.40 | $64.00 | $4.00 | $0.40 | $24.00 |
| gpt-realtime-1.5 | $32.00 | $0.40 | $64.00 | $4.00 | $0.40 | $16.00 |
| gpt-realtime | $32.00 | $0.40 | $64.00 | $4.00 | $0.40 | $16.00 |
| gpt-realtime-2.1-mini | $10.00 | $0.30 | $20.00 | $0.60 | $0.06 | $2.40 |
| gpt-realtime-mini | $10.00 | $0.30 | $20.00 | $0.60 | $0.06 | $2.40 |
| gpt-audio-1.5 | $32.00 | - | $64.00 | $2.50 | - | $10.00 |
| gpt-audio | $32.00 | - | $64.00 | $2.50 | - | $10.00 |
| gpt-audio-mini | $10.00 | - | $20.00 | $0.60 | - | $2.40 |
| gpt-4o-mini-tts | - | - | $12.00 | $0.60 | - | - |
The realtime models also take image input: $5.00 per million tokens on the full-size models and $0.80 on the minis.1 The older text-to-speech models bill by character: tts-1 costs $15.00 and tts-1-hd $30.00 per million characters.1
Transcription and translation:1
| Model | Use | Listed cost per minute | Token rates ($/1M, input / output) |
|---|---|---|---|
| gpt-4o-mini-transcribe | Transcription | $0.003 (estimated) | $1.25 / $5.00 |
| gpt-transcribe | Transcription | $0.0045 | - |
| gpt-4o-transcribe | Transcription | $0.006 (estimated) | $2.50 / $10.00 |
| gpt-4o-transcribe-diarize | Transcription with diarization | $0.006 (estimated) | $2.50 / $10.00 |
| Whisper | Transcription | $0.006 | - |
| gpt-live-transcribe | Live transcription | $0.017 | - |
| gpt-realtime-whisper | Live transcription | $0.017 | - |
| gpt-realtime-translate | Live translation | $0.034 | - |
OpenAI’s product page converts the per-minute models to per-second rates: $0.00083 for GPT-Live-1, $0.00057 for GPT-Realtime-Translate, $0.00028 for GPT-Live-Transcribe and $0.00008 for GPT-Transcribe.2 The token-billed transcription models carry OpenAI’s own per-minute estimate, so a real bill depends on the audio.
What do OpenAI’s image generation models cost?
OpenAI’s GPT Image 2.5 models, gpt-image-2.5-sunburst and gpt-image-2.5-flare, both cost $8.00 per million image input tokens, $2.00 cached and $30.00 per million image output tokens, plus $5.00 per million text input tokens. The cheapest image model is gpt-image-1-mini at $2.50 image input and $8.00 image output. The most expensive is gpt-image-1 at $10.00 and $40.00.1
| Model | Image input | Image cached input | Image output | Text input | Text cached input | Text output | Batch image input / output |
|---|---|---|---|---|---|---|---|
| gpt-image-2.5-sunburst | $8.00 | $2.00 | $30.00 | $5.00 | $1.25 | - | not listed |
| gpt-image-2.5-flare | $8.00 | $2.00 | $30.00 | $5.00 | $1.25 | - | not listed |
| gpt-image-2 | $8.00 | $2.00 | $30.00 | $5.00 | $1.25 | - | $4.00 / $15.00 |
| gpt-image-1.5 | $8.00 | $2.00 | $32.00 | $5.00 | $1.25 | $10.00 | $4.00 / $16.00 |
| chatgpt-image-latest | $8.00 | $2.00 | $32.00 | $5.00 | $1.25 | $10.00 | $4.00 / $16.00 |
| gpt-image-1-mini | $2.50 | $0.25 | $8.00 | $2.00 | $0.20 | - | $1.25 / $4.00 |
| gpt-image-1 | $10.00 | $2.50 | $40.00 | $5.00 | $1.25 | - | $5.00 / $20.00 |
All prices per 1M tokens.1 Cached input rates for GPT Image 2 and GPT Image 2.5 apply only to images generated with the Responses API.1 OpenAI’s product page shows a Batch option at 50% off for image models, but the docs Batch table has no rows for the two GPT Image 2.5 models.2,1
Images sent to a language model are billed differently. OpenAI says text models price image tokens at standard text token rates, while GPT Image and gpt-realtime use a separate image token rate, and gpt-4.1-mini, gpt-4.1-nano and o4-mini convert images into tokens differently.2
What do embeddings, moderation and built-in tools cost?
OpenAI’s embedding models cost $0.02 per million tokens for text-embedding-3-small, $0.10 for text-embedding-ada-002 and $0.13 for text-embedding-3-large, and omni-moderation-latest is free. Built-in tools add per-call and storage charges on top of token costs. Web search costs $10.00 per 1,000 calls, plus the search content tokens at the model’s rates.1
| Tool | OpenAI’s listed price |
|---|---|
| Web search, all models | $10.00 per 1,000 calls, plus search content tokens at model rates |
| Web search preview, reasoning models | $10.00 per 1,000 calls, plus search content tokens at model rates |
| Web search preview, non-reasoning models | $25.00 per 1,000 calls; search content tokens free |
| Containers (Hosted Shell and Code Interpreter) | 1 GB $0.03, 4 GB $0.12, 16 GB $0.48, 64 GB $1.92 per 20-minute session per container |
| File search storage | $0.10 per GB per day (1 GB free) |
| File search tool calls | $2.50 per 1,000 calls |
| ChatKit file and image upload storage | $0.10 per GB-day after 1 GB free per account per month |
Tokens used by built-in tools are billed at the chosen model’s per-token rates. For gpt-4o-mini and gpt-4.1-mini with the non-preview web search tool, search content is billed as a fixed block of 8,000 input tokens per call.1 At listed rates, that block adds $0.0012 per call on gpt-4o-mini and $0.0032 on gpt-4.1-mini, beside the $0.01 call fee.1
Why do other published OpenAI prices disagree?
Glamdring Research compared three OpenAI pricing pages and two third-party tables captured on 28 September 2026. The disagreements come from stale prose, copying errors and different units. OpenAI’s docs rate card, captured at two docs addresses six hours apart, gave the same GPT-6 rates both times.1,3 When a third-party figure differs, the docs page decides.
- BenchLM contradicts its own table. Its table lists GPT-5.6 Sol at $4 input and $20 output, matching OpenAI. Its buying guide puts GPT-5.6 Sol and GPT-5.5 together at “$5/$30”.4 OpenAI lists gpt-5.5 at $5.00 and $30.00 and gpt-5.6-sol at a promotional $4.00 and $20.00.1
- SpendHound differs on one long-context rate. It lists GPT-5.6 Luna’s long-context output at $1.90.5 OpenAI lists $1.80.1
- SpendHound’s headline figures are a different unit. Its $44,318 and $318,506 are average annual spend among its SMB and enterprise customers, drawn from its own spend data.5 They describe contract totals and say nothing about per-token rates.
- OpenAI’s own pages differ. The openai.com product page says web search costs $10.00 per 1,000 calls and that search content tokens are free.2 The docs page bills those tokens at model rates, except on web search preview with non-reasoning models.1 The product page’s container line still describes pricing “Starting March 31, 2026” as a future change.2
- Dates move the numbers. Priority processing became Fast mode on July 30, 2026, GPT-5.6 Sol’s promotional rate runs at least through November 21, 2026, and gpt-rosalind-research billing starts on October 5, 2026.1 A table captured on either side of those dates will differ from this one.
What the ranking cannot tell you
The ranking orders list prices. It can’t say which model finishes a given job at the lowest cost. A model with a lower input rate can cost more per completed task if it writes more output tokens or needs more attempts. Use one real workflow, load case or unit model to test the claim.
The captured pages also leave out several things that change a bill:
- Currency and tax. OpenAI’s pages show a dollar sign with no currency code or tax statement.1
- Negotiated terms. OpenAI sells Scale Tier and Reserved Capacity through its sales team for larger workloads, and the pages publish no rates for them.2
- Spend limits. OpenAI lets an account set a monthly budget, but says “There may be a delay in enforcing the limit, and you are responsible for any overage incurred.”2
- Quality. Nothing here measures output quality, speed or context capacity.
For other providers’ list prices on the same per-million-token basis, see our Claude API pricing reference and our review of OpenAI API alternatives.
Frequently asked questions
Is the OpenAI API free or paid?
The OpenAI API is paid. OpenAI bills API use per token at the listed input and output rates, and it bills Playground use the same way.2 One model is listed as free: omni-moderation-latest.1 The pricing pages captured for this ranking list no general free tier.
How much is 1 million tokens on OpenAI?
One million tokens costs between $0.05 and $150.00 as input, and between $0.40 and $600.00 as output, depending on the model.1 At Standard short-context rates, one million input tokens plus one million output tokens costs $0.60 on gpt-6-luna, $12.00 on gpt-6-sol and $60.00 on gpt-6-astra. Batch halves the gpt-6-sol figure to $6.00.1
Does a $20 ChatGPT Plus subscription include API access?
No. ChatGPT Plus costs $20 a month,6 and OpenAI states that its APIs are billed separately from ChatGPT Plus, Business, Enterprise and Edu.2 A Plus subscription and an API account are two separate purchases.
Can you still fine-tune OpenAI models, and what does it cost?
Only existing fine-tuning users can still train models. OpenAI is winding down its fine-tuning platform: new users can’t access it, existing users can create training jobs “for the coming months”, and fine-tuned models stay available for inference until their base models are deprecated.1 A fine-tuned gpt-4.1-2025-04-14 costs $25.00 per 1M training tokens and $3.00 input, $0.75 cached input and $12.00 output per 1M tokens at inference. Training o4-mini-2025-04-16 costs $100.00 an hour.1
Sources checked
Company-owned pages establish what a company says. They do not prove a market conclusion. Each source is dated so readers can judge each claim.
