OpenAI GPT-5 API pricing: every model, September 2026
gpt-5-nano is the cheapest GPT-5 model on the OpenAI API at $0.05 input and $0.40 output per million tokens; gpt-5.5-pro and gpt-5.4-pro are the dearest at $30 and $180. Batch and Flex halve the standard rate, and Fast mode doubles it on most models.
The cheapest GPT-5 model on the OpenAI API is gpt-5-nano, at $0.05 per million input tokens and $0.40 per million output tokens on OpenAI’s developer pricing page, checked 28 September 2026.1 The dearest are gpt-5.5-pro and gpt-5.4-pro, at $30 input and $180 output.1 That’s a 600-fold spread in input price across 20 models, so the choice of model moves a bill further than any discount tier can. This ranking sits in our model infrastructure research.
TL;DR
- Batch and Flex each charge half the standard rate. Batch covers 16 of the 20 models and Flex covers 14; gpt-5-pro and gpt-5.2-pro appear in Batch but not Flex.1
- Fast mode, which OpenAI called priority processing until 30 July 2026, doubles the rate on nine of the 11 GPT-5 models it lists. It costs 2.5 times standard on gpt-5.5 and 1.8 times on gpt-5-mini.1
- Cached input costs a tenth of the input rate on all 16 models that publish one. The four Pro models publish no cached rate.1
- A newer version can cost less. gpt-5.6-luna charges $0.20 input and $1.20 output against gpt-5-mini’s $0.25 and $2.00.1
- Input price alone can mislead. gpt-5.6-terra costs more than gpt-5.2 on input and less on output, so it’s the cheaper of the two whenever a prompt is under eight times the length of its reply.1
- Seven models publish a long-context rate, at twice the standard input price and 1.5 times the output price.1
How this ranking was built
Every figure comes from OpenAI’s API pricing documentation at developers.openai.com, captured at 21:12 AEST on 28 September 2026 with every “All models” toggle open.1 The ranking covers every GPT-5 family model’s API price per million tokens from OpenAI’s pricing pages, checked 28 September 2026, ordered by input price.
The ranked field is OpenAI’s standard short-context Input price, in dollars per 1M tokens, cheapest first. A model qualifies when its API name begins with gpt-5. With its toggle open, the standard Flagship table lists 39 text models, and 16 of them are GPT-5 models.1 The Cyber table adds gpt-5.6-cyber and gpt-5.5-cyber. Its third row, gpt-5.6-sol, repeats the Flagship price and is counted once.1 The Specialized table adds gpt-5.3-codex and gpt-5-search-api.1 That gives 20 models.
Left out: the three GPT-6 models, chat-latest (which OpenAI doesn’t tie to a version number), and the two gpt-daybreak aliases, which point to gpt-5.6-sol and gpt-5.6-cyber.1 No image, audio or transcription model carries a GPT-5 name. Ties share a rank and keep OpenAI’s page order. Figures appear exactly as OpenAI prints them, with a dash where OpenAI publishes no rate. OpenAI quotes every rate with a dollar sign, and the captured page names no currency code or tax treatment. The full research standard is on our methodology page.
OpenAI’s platform.openai.com pricing address, captured at 15:05 AEST the same day, shows only the three GPT-6 models in its Flagship tables until “All models” is opened.2 The GPT-5 rows it does show by default, gpt-5.6-sol in the Cyber table and gpt-5.3-codex in the Specialized table, match the expanded capture.2
What does each GPT-5 model cost per million tokens?
Glamdring Research ranked 20 GPT-5 models by standard input price on the OpenAI API, checked 28 September 2026. gpt-5-nano is cheapest at $0.05 input, $0.005 cached input and $0.40 output per million tokens. gpt-5.5-pro and gpt-5.4-pro share the top price at $30 input and $180 output, with no cached rate.1 All figures are OpenAI’s standard short-context rates.
| Rank | Model | Page section | Input ($ per 1M tokens) | Cached input ($ per 1M) | Output ($ per 1M) |
|---|---|---|---|---|---|
| 1 | gpt-5-nano | Flagship | $0.05 | $0.005 | $0.40 |
| 2= | gpt-5.6-luna | Flagship | $0.20 | $0.02 | $1.20 |
| 2= | gpt-5.4-nano | Flagship | $0.20 | $0.02 | $1.25 |
| 4 | gpt-5-mini | Flagship | $0.25 | $0.025 | $2.00 |
| 5 | gpt-5.4-mini | Flagship | $0.75 | $0.075 | $4.50 |
| 6= | gpt-5.1 | Flagship | $1.25 | $0.125 | $10.00 |
| 6= | gpt-5 | Flagship | $1.25 | $0.125 | $10.00 |
| 6= | gpt-5-search-api | Specialized (Search) | $1.25 | $0.125 | $10.00 |
| 9= | gpt-5.2 | Flagship | $1.75 | $0.175 | $14.00 |
| 9= | gpt-5.3-codex | Specialized (Codex) | $1.75 | $0.175 | $14.00 |
| 11 | gpt-5.6-terra | Flagship | $2.00 | $0.20 | $12.00 |
| 12 | gpt-5.4 | Flagship | $2.50 | $0.25 | $15.00 |
| 13 | gpt-5.6-sol | Flagship (promotional) | $4.00 | $0.40 | $20.00 |
| 14 | gpt-5.5 | Flagship | $5.00 | $0.50 | $30.00 |
| 15= | gpt-5.6-cyber | Cyber | $12.50 | $1.25 | $75.00 |
| 15= | gpt-5.5-cyber | Cyber | $12.50 | $1.25 | $75.00 |
| 17 | gpt-5-pro | Flagship | $15.00 | - | $120.00 |
| 18 | gpt-5.2-pro | Flagship | $21.00 | - | $168.00 |
| 19= | gpt-5.5-pro | Flagship | $30.00 | - | $180.00 |
| 19= | gpt-5.4-pro | Flagship | $30.00 | - | $180.00 |
The output rate is where the generations part ways. Every gpt-5, gpt-5.1 and gpt-5.2 model charges eight times its input rate for output, Pro, mini and nano versions included. From gpt-5.4 onwards most models charge six times, and gpt-5.6-sol charges five.1 A workload that writes long replies gains more from the newer models than the input column suggests.
Several rows share one price. gpt-5 and gpt-5.1 are identical, and gpt-5-search-api matches both. gpt-5.3-codex matches gpt-5.2 on every column.1 gpt-5.6-sol carries a flag of its own: OpenAI says its promotional pricing “is available at least through November 21, 2026”.1
What do cached input and the Batch API cost for GPT-5 models?
Cached input costs 10% of the standard input rate on every GPT-5 model that lists one, and the Batch API halves input, cached input and output. On gpt-5, cached input drops from $1.25 to $0.125 per million tokens. In Batch it costs $0.625 input, $0.0625 cached input and $5.00 output.1 OpenAI says Batch runs tasks asynchronously over 24 hours.3
| Model | Batch input ($ per 1M) | Batch cached input ($ per 1M) | Batch output ($ per 1M) | Flex |
|---|---|---|---|---|
| gpt-5-nano | $0.025 | $0.0025 | $0.20 | Same as Batch |
| gpt-5.6-luna | $0.10 | $0.01 | $0.60 | Same as Batch |
| gpt-5.4-nano | $0.10 | $0.01 | $0.625 | Same as Batch |
| gpt-5-mini | $0.125 | $0.0125 | $1.00 | Same as Batch |
| gpt-5.4-mini | $0.375 | $0.0375 | $2.25 | Same as Batch |
| gpt-5.1 | $0.625 | $0.0625 | $5.00 | Same as Batch |
| gpt-5 | $0.625 | $0.0625 | $5.00 | Same as Batch |
| gpt-5.2 | $0.875 | $0.0875 | $7.00 | Same as Batch |
| gpt-5.6-terra | $1.00 | $0.10 | $6.00 | Same as Batch |
| gpt-5.4 | $1.25 | $0.13 | $7.50 | Same as Batch |
| gpt-5.6-sol | $2.00 | $0.20 | $10.00 | Same as Batch |
| gpt-5.5 | $2.50 | $0.25 | $15.00 | Same as Batch |
| gpt-5-pro | $7.50 | - | $60.00 | Not listed |
| gpt-5.2-pro | $10.50 | - | $84.00 | Not listed |
| gpt-5.5-pro | $15.00 | - | $90.00 | Same as Batch |
| gpt-5.4-pro | $15.00 | - | $90.00 | Same as Batch |
Four models have no Batch price. gpt-5.3-codex and gpt-5-search-api sit in a Specialized section that offers only Standard and Fast mode tabs, and the two Cyber models have a single price table.1 OpenAI prints gpt-5.4’s Batch cached rate as $0.13, a rounded half of its $0.25 standard rate.1
The GPT-5.6 models list a cache-write charge. It costs 1.25 times input, at $5.00 per million tokens on gpt-5.6-sol, $2.50 on gpt-5.6-terra and $0.25 on gpt-5.6-luna. gpt-5.6-cyber lists $15.625.1 The other GPT-5 rows show a dash or have no cache-write column. On a workload that rebuilds its cache often, that write charge narrows the gap between a GPT-5.6 model and an older one.
What do Flex processing and Fast mode cost?
Flex costs the same as Batch, half the standard rate, and Fast mode costs about double. Fast mode is the name OpenAI has used since 30 July 2026 for what it called priority processing.1 On gpt-5, Flex charges $0.625 input and $5.00 output per million tokens, and Fast mode charges $2.50 and $20.00.1 OpenAI says Flex gives “lower costs for requests in exchange for slower response times and occasional resource unavailability”.3
| Model | Fast mode input ($ per 1M) | Fast mode cached input ($ per 1M) | Fast mode output ($ per 1M) | Multiple of standard |
|---|---|---|---|---|
| gpt-5.6-luna | $0.40 | $0.04 | $2.40 | 2 |
| gpt-5-mini | $0.45 | $0.045 | $3.60 | 1.8 |
| gpt-5.4-mini | $1.50 | $0.15 | $9.00 | 2 |
| gpt-5.1 | $2.50 | $0.25 | $20.00 | 2 |
| gpt-5 | $2.50 | $0.25 | $20.00 | 2 |
| gpt-5.2 | $3.50 | $0.35 | $28.00 | 2 |
| gpt-5.3-codex | $3.50 | $0.35 | $28.00 | 2 |
| gpt-5.6-terra | $4.00 | $0.40 | $24.00 | 2 |
| gpt-5.4 | $5.00 | $0.50 | $30.00 | 2 |
| gpt-5.6-sol | $8.00 | $0.80 | $40.00 | 2 |
| gpt-5.5 | $12.50 | $1.25 | $75.00 | 2.5 |
Nine GPT-5 models have no Fast mode price: gpt-5-nano, gpt-5.4-nano, the four Pro models, gpt-5-search-api and both Cyber models.1 Flex leaves out gpt-5-pro, gpt-5.2-pro, the two Specialized models and the Cyber pair.1 Requests select Fast mode with either service_tier: "priority" or service_tier: "fast".1
The multiple column is our arithmetic on OpenAI’s rates. gpt-5.5 is the outlier at 2.5 times, which puts its Fast mode output at $75 per million tokens, the same as the Cyber models’ standard rate.1 Microsoft’s Azure price list shows the same doubled rates under the label “Priority Processing” for gpt-5.6-sol, at $8 input and $40 output, and the same $12.50 and $75 for gpt-5.5.4
How much more does a long prompt cost?
A prompt over the long-context boundary costs twice the standard input rate and 1.5 times the output rate on the seven GPT-5 models that publish long-context prices. gpt-5.4 rises from $2.50 to $5.00 per million input tokens and from $15.00 to $22.50 output.1 The other 13 GPT-5 models list a single rate with no long-context column.
| Model | Short-context input | Long-context input | Short-context output | Long-context output |
|---|---|---|---|---|
| gpt-5.6-luna | $0.20 | $0.40 | $1.20 | $1.80 |
| gpt-5.6-terra | $2.00 | $4.00 | $12.00 | $18.00 |
| gpt-5.4 | $2.50 | $5.00 | $15.00 | $22.50 |
| gpt-5.6-sol | $4.00 | $8.00 | $20.00 | $30.00 |
| gpt-5.5 | $5.00 | $10.00 | $30.00 | $45.00 |
| gpt-5.5-pro | $30.00 | $60.00 | $180.00 | $270.00 |
| gpt-5.4-pro | $30.00 | $60.00 | $180.00 | $270.00 |
Prices are dollars per 1M tokens at standard rates. Cached input and cache writes double too: gpt-5.6-sol’s cached rate goes from $0.40 to $0.80 and its cache write from $5.00 to $10.00.1 Fast mode publishes long-context rates only for the three GPT-5.6 models, and gpt-5.5-pro has no long-context Batch or Flex price.1
The developer page doesn’t print the boundary beside the GPT-5 rows. OpenAI’s business pricing page says its quoted pricing “reflects standard processing rates for context lengths under 272K”, on a page that lists only GPT-6 models.3 Microsoft labels its GPT-5.4 tiers “<272k context length” and “>272k context length”.4 Price Per Token’s GPT-5 page uses a different cut-off and shows prices “for prompts ≤ 200k tokens”.5 Confirm the boundary with OpenAI before budgeting long-document work.
What does one typical request cost on each model?
A request with a 10,000-token prompt and a 1,000-token reply costs $0.0225 on gpt-5 at standard rates, $0.0009 on gpt-5-nano and $0.48 on gpt-5.5-pro. That’s Glamdring Research’s arithmetic on OpenAI’s rates, checked 28 September 2026: input tokens times the input rate, plus output tokens times the output rate, divided by one million.1
| Model | Standard, per 1,000 requests | Batch, per 1,000 requests | Fast mode, per 1,000 requests |
|---|---|---|---|
| gpt-5-nano | $0.90 | $0.45 | - |
| gpt-5.6-luna | $3.20 | $1.60 | $6.40 |
| gpt-5.4-nano | $3.25 | $1.625 | - |
| gpt-5-mini | $4.50 | $2.25 | $8.10 |
| gpt-5.4-mini | $12.00 | $6.00 | $24.00 |
| gpt-5 | $22.50 | $11.25 | $45.00 |
| gpt-5.2 | $31.50 | $15.75 | $63.00 |
| gpt-5.6-terra | $32.00 | $16.00 | $64.00 |
| gpt-5.4 | $40.00 | $20.00 | $80.00 |
| gpt-5.6-sol | $60.00 | $30.00 | $120.00 |
| gpt-5.5 | $80.00 | $40.00 | $200.00 |
| gpt-5-pro | $270.00 | $135.00 | - |
| gpt-5.5-pro | $480.00 | $240.00 | - |
The shape of the request changes the order. At 10,000 tokens in and 1,000 out, gpt-5.2 edges gpt-5.6-terra by 50 cents per 1,000 requests. Halve the prompt to 5,000 tokens and gpt-5.6-terra wins, at $22.00 against $22.75. The two cost the same when the prompt is exactly eight times the reply.1 Run the sum on your own token counts before choosing between neighbouring rows.
Caching moves the bill further. If 8,000 of the 10,000 prompt tokens hit the cache on gpt-5, the request costs $13.50 per 1,000 instead of $22.50, a 40% cut. Run the same request through Batch and it falls to $6.75, because Batch prints its own cached rate of $0.0625.1 On the GPT-5.6 models, the cache-write charge eats into that saving, and the price page doesn’t say how often a write recurs.
Our Claude API pricing ranking prices the same 10,000-token request on Anthropic’s models from Anthropic’s page on the same date, so the two tables compare line by line. A production bill also carries retries, tool calls and review time. Our explainer on what controls AI inference cost works through those.
Why do other published GPT-5 prices disagree?
Most disagreement comes from rounding, reseller uplifts and model names absent from OpenAI’s price page. Microsoft’s Azure page lists gpt-5 cached input at $0.13 where OpenAI prints $0.125, and gpt-5.4-mini cached input at $0.08 where OpenAI prints $0.075.4,1 The base input and output rates match.
Azure’s Data Zone deployments add 10%. gpt-5.6-sol costs $4 input on Azure Global and $4.40 in a Data Zone.4 Azure also lists GPT-5 Codex, GPT-5 chat and GPT-5.1-codex-max at $1.25 input and $10 output.4 None of those names appears on OpenAI’s developer price page in our capture. Price Per Token’s GPT-5 page agrees with OpenAI on gpt-5 itself, at $1.25 input and $10.00 output, and lists GPT-5 Codex and GPT-5 Chat beside it.5,1
OpenAI’s own pages are the third cause. Its business pricing page at openai.com lists GPT-6 Astra, Sol and Luna and no GPT-5 model.3 The developer page’s default view includes GPT-5.6 Sol, GPT-5.6 Cyber and GPT-5.3 Codex.2 Expanding “All models” exposes the rest of the GPT-5 family.1
What changes are dated on OpenAI’s price page?
Two dated notes touch GPT-5 prices. OpenAI renamed priority processing to Fast mode on 30 July 2026, and both service_tier values still work.1 OpenAI also says gpt-5.6-sol’s promotional pricing lasts at least through 21 November 2026, and it doesn’t say what rate follows.1 This is Glamdring’s first capture of the page, so no earlier GPT-5 figures are compared here.
The bigger shift sits one family up. GPT-6 Sol lists at $2.00 input and $10.00 output, half the $4.00 and $20.00 that gpt-5.6-sol charges. GPT-6 Luna lists at $0.10 and $0.50 against gpt-5.6-luna’s $0.20 and $1.20.1 The decision changes when a GPT-6 model handles the same workload, because the per-token rate falls by half or more. The price page says nothing about whether it does.
What the ranking cannot tell you
This ranking orders list prices. It says nothing about capability, speed or output quality. A cheaper rate doesn’t guarantee a cheaper finished task, because a model that needs two attempts can cost more than a dearer one that succeeds first time.
Several charges sit outside the table. Regional processing endpoints add 10% for eligible models released on or after 5 March 2026, and the price page gives no release dates, so it doesn’t settle which GPT-5 models carry that uplift.1 FedRAMP endpoints add 10% over standard rates.1 Web search costs $10.00 per 1,000 calls, plus search content tokens at the model’s rates.1 OpenAI models on Amazon Bedrock bill through AWS, at prices OpenAI says match its direct prices in commercial regions.1
The page lists Cyber model prices without stating who can buy them. It names no currency code and no tax treatment. These are OpenAI’s published list rates, and enterprise terms such as Scale Tier and Reserved Capacity go through OpenAI’s sales team.3 New price checks across model infrastructure go out in our free email briefing.
Frequently asked questions
How much do 1,000 tokens cost on GPT-5?
On gpt-5, 1,000 input tokens cost $0.00125 and 1,000 output tokens cost $0.01 at standard rates. Divide any per-million price by 1,000 to get the per-1,000 figure.1 Across the family, 1,000 input tokens run from $0.00005 on gpt-5-nano to $0.03 on gpt-5.5-pro and gpt-5.4-pro.1
Is the OpenAI API free or paid?
Paid, per token. OpenAI bills Playground use the same way as regular API use.3 The developer price page lists one free model, omni-moderation-latest.1 You can set a monthly budget in billing settings, though OpenAI says enforcement may lag and you’re responsible for any overage.3
Does a $20 ChatGPT Plus plan include API access?
No. ChatGPT Plus lists at $20 a month.6 OpenAI says its APIs are billed separately from ChatGPT Plus, Business, Enterprise and Edu.3 A Plus subscriber who calls gpt-5 from code pays the per-token rates in the table above, billed apart from the subscription.
How much does the ChatGPT API cost?
The model OpenAI files under “ChatGPT” on its API price page is chat-latest, at $5.00 input, $0.50 cached input and $30.00 output per million tokens.1 The page doesn’t say which model version chat-latest points to. For a fixed version, the GPT-5 models above run from $0.05 to $30 per million input tokens.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.
