amBrain

OpenAI API vs Claude API: prices, data terms, retention and model retirement compared

Published Sep 23, 2026Facts checked: Sep 23, 2026

Short answer

On published terms the two APIs have a great deal in common. Both price per million tokens. Both halve the price of batch work: Anthropic advertises "Save 50% with batch processing", and on OpenAI's table the Batch column is exactly half the Standard column for every model listed. Both say API data is not used for training by default, and both leave you owning your outputs.[1][2][4][5][6] They differ on retirement notice, at least 6 months for OpenAI's generally available models against at least 60 days at Anthropic[3][8], on regional processing and its surcharges, and on channels: Anthropic offers Claude "in the cloud you already use, including Amazon Web Services, Google Cloud, or Microsoft Foundry".[9] Output quality is not compared here; test both on your own data.

On this page

Side by side

What OpenAI and Anthropic publish for their APIs, read on 2026-09-23
DimensionOpenAI APIClaude API (Anthropic)
Top-tier price per 1M tokensgpt-6-astra: $10.00 input, $1.00 cached input, $50.00 output; gpt-5.5: $5.00, $0.50, $30.00; gpt-5.5-pro: $30.00 and $180.00[1]Opus 5.5: "$4 / MTok" input, "$20 / MTok" output, cache read $0.20, cache write $5; Fable 5.1: $10 input, $50 output. Opus 5 at $5 and $25 now sits under the "Legacy models" heading[5]
Mid-tier price per 1M tokensgpt-6-sol: $2.00 input, $0.20 cached input, $10.00 output; gpt-5.1: $1.25, $0.125, $10.00[1]Sonnet 5: $2 input, $10 output[5]
Small-tier price per 1M tokensgpt-6-luna: $0.10 input, $0.01 cached input, $0.50 output; gpt-5-mini: $0.25 input, $2.00 output; gpt-5-nano: $0.05 input, $0.40 output[1]Haiku 4.5: $1 input, $5 output[5]
Batch processingNo percentage is stated; on the published table the Batch column is exactly half the Standard column for every model listed[1]"Save 50% with batch processing"[5]
Faster tier"Priority processing was renamed Fast mode on July 30, 2026"[1]"Get up to 2.5x faster speeds with fast mode for Opus 5.5 at 2x standard pricing"[5]
Training on API data"data sent to the OpenAI API is not used to train or improve OpenAI models (unless you explicitly opt in to share data with us)"[2]"Anthropic may not train models on Customer Content from Services"[6]
Ownership of inputs and outputs"Customer: (a) retains all ownership rights in Input; and (b) owns all Output"[4]"Customer (a) retains all rights to its Inputs, and (b) owns its Outputs"[6]
Default retentionAbuse-monitoring logs retained for up to 30 days[2]Inputs and outputs deleted from the back end within 30 days; if a chat is flagged, "We retain inputs and outputs for up to 2 years and trust and safety classification scores for up to 7 years"[7]
Zero data retentionZero Data Retention or Modified Abuse Monitoring for eligible customers[2]Zero data retention by agreement[7]
Regional processingData-residency endpoints in the United States, "Europe (EEA + Switzerland)", Australia, Canada, Japan, India, Singapore, South Korea, the United Kingdom and the United Arab Emirates, where "Selecting the United Arab Emirates region requires additional approval"[2]; a 10% uplift for models released on or after March 5, 2026[1]US-only inference priced at 1.1x[5]
Retirement noticeGenerally available models at least 6 months; specialised variants at least 3 months; preview models a short period "such as 2 weeks"[3]At least 60 days' notice before retirement of publicly released models; each active model shows a "not sooner than" date[8]
Minimum term and terminationNo minimum commitment stated on the pricing page[1]; termination terms not compared hereNo minimum term; either party may terminate for convenience, Anthropic on 30 days' notice[6]
Other channels for the same modelsOpenAI models are also sold by Microsoft: a "model sold by Azure" "includes Azure OpenAI models", and Microsoft states the prompts are not available to OpenAI[10]"Or use Claude in the cloud you already use, including Amazon Web Services, Google Cloud, or Microsoft Foundry"[9]. Anthropic's retirement dates cover "the Claude API, Claude Platform on AWS, and Microsoft Foundry", while "Partner-operated platforms (Amazon Bedrock and Google Cloud) set their own retirement schedules"[8]

Prices change often. Read both pricing pages on the day you decide, and compare the models you would use, not the top of each list.

What the numbers mean for a document or ticket workload

For extraction and classification, input tokens usually outnumber output tokens many times over, because each request carries a page or a ticket and returns a short record. That makes input price, cached-input price and batch discounts the lines to compare first. Output price matters more for drafting and summarisation.

Retirement notice turns into engineering work. Every retirement means scoring the replacement model on your own evaluation set before the date. With OpenAI's six months for generally available models[3] that is planned work; with Anthropic's 60 days[8] it has to be a standing routine. A longer notice period does not remove the work, it moves the deadline.

Regional processing is priced on both sides: OpenAI adds 10% for data-residency endpoints on models released on or after March 5, 2026[1], and Anthropic prices US-only inference at 1.1x[5]. If residency is a requirement, compare prices with the uplift included.

Choose OpenAI's API when

  1. 1.You need processing in a specific region, such as India, Singapore or the United Arab Emirates, which OpenAI lists among its data-residency endpoints (with the caveat that "Selecting the United Arab Emirates region requires additional approval"[2]), and you accept the residency uplift[1]. The regional option on Anthropic's pricing page is US-only inference at 1.1x[5]; check its current list for your region.
  2. 2.The workload is high-volume and simple, such as routing or short classification, where the smallest listed tier, gpt-5-nano at $0.05 input and $0.40 output per million tokens[1], passes your evaluation.
  3. 3.Your change process needs long notice before a model disappears: at least 6 months for generally available models[3].

Choose the Claude API when

  1. 1.You want the same models under a cloud contract you already hold: Anthropic offers Claude "in the cloud you already use, including Amazon Web Services, Google Cloud, or Microsoft Foundry", as well as directly[9]. Check the retirement schedule with it: Anthropic's dates cover Microsoft Foundry, while "Partner-operated platforms (Amazon Bedrock and Google Cloud) set their own retirement schedules"[8].
  2. 2.You want a contract you can leave for convenience: Anthropic's commercial terms set no minimum term and let either party terminate for convenience, Anthropic on 30 days' notice[6].
  3. 3.You need US-only inference as a contracted option rather than a region choice, priced at 1.1x[5].

Choose neither when

  1. 1.Documents may not leave infrastructure you operate, even under contract. An open-weight model served with vLLM, which ships an "OpenAI-compatible API server, plus Anthropic Messages API and gRPC support"[11], keeps client code close to either API while the model runs inside your perimeter.
  2. 2.Your organisation already runs its workloads on one cloud and needs private networking and the cloud's own data terms. A managed service such as Amazon Bedrock, or a model sold by Azure in Microsoft Foundry[10], may fit better; see AWS Bedrock alternatives.
  3. 3.The task is page-level OCR on standard forms. Per-page document-AI services can cost less than one LLM call per page, so price the same sample of pages both ways before you commit to either API.

Keeping the choice reversible

Put one interface between your application and any model provider: the prompt, the output schema, the model name and the provider are configuration, and every call is logged with the model version that answered. Keep an evaluation set of your own documents with expected outputs. Then switching provider, or adding a second one for failover, is a scored change rather than a rewrite. The deployment choice for a regulated perimeter is covered in the article on choosing an LLM deployment for a closed perimeter, and the pipeline around the model in the article on an LLM pipeline for claims, tickets and KYC files.

About amBrain

  • We have taken an LLM integration to production inside a client's FinTech perimeter: extracting and normalising unstructured broker and venue notices — corporate actions, instrument and margin changes — into structured records the trading system consumes.

Frequently asked questions

It depends on the tier and the mix of input and output tokens. At the top of each list OpenAI prices gpt-6-astra at $10.00 input and $50.00 output per million tokens, while Anthropic prices Opus 5.5 at "$4 / MTok" and "$20 / MTok"[1][5]. One tier down, gpt-6-sol is $2.00 and $10.00 against Sonnet 5 at $2 and $10[1][5]. Anthropic advertises "Save 50% with batch processing", and on OpenAI's table the Batch column is exactly half the Standard column for every model listed[5][1].

Both say no by default. OpenAI states that "data sent to the OpenAI API is not used to train or improve OpenAI models (unless you explicitly opt in to share data with us)"[2], and Anthropic's commercial terms state that "Anthropic may not train models on Customer Content from Services"[6].

OpenAI keeps abuse-monitoring logs for up to 30 days, with Zero Data Retention or Modified Abuse Monitoring for eligible customers[2]. Anthropic deletes inputs and outputs from the back end within 30 days; if a chat is flagged, "We retain inputs and outputs for up to 2 years and trust and safety classification scores for up to 7 years", and zero data retention is available by agreement[7].

OpenAI gives at least 6 months for generally available models, at least 3 months for specialised variants and a short period "such as 2 weeks" for previews[3]. Anthropic gives at least 60 days for publicly released models and shows a "not sooner than" date for each active model; its dates apply to "the Claude API, Claude Platform on AWS, and Microsoft Foundry", while "Partner-operated platforms (Amazon Bedrock and Google Cloud) set their own retirement schedules"[8].

Yes, if the application talks to the model through one interface and you keep an evaluation set. vLLM also ships an "OpenAI-compatible API server, plus Anthropic Messages API and gRPC support"[11], so the same client code can move to a self-hosted model.

Disclosure: this page is published by amBrain. Every vendor fact on it links to the page it was read from on 2026-09-23, and both providers change prices and model line-ups without notice. No vendor named here reviewed or endorsed this page. OpenAI, Anthropic, Claude, Microsoft, Azure, Microsoft Foundry, Amazon Web Services, Amazon Bedrock, Google Cloud and vLLM are trademarks of their owners and are used only to identify the products discussed.

Sources

  1. [1]OpenAI, Pricing (flagship and small-model tables with Standard, Batch and cached-input columns; Fast mode note; data-residency uplift). Vendor's own page · checked Sep 23, 2026
  2. [2]OpenAI, Your data (training opt-in, 30-day abuse-monitoring logs, Zero Data Retention and Modified Abuse Monitoring, list of data-residency regions). Official documentation · checked Sep 23, 2026
  3. [3]OpenAI, Deprecations (notice periods by model class). Official documentation · checked Sep 23, 2026
  4. [4]OpenAI, OpenAI Services Agreement (version ONLINE v.010126; §4.1 ownership of Input and Output). Vendor's published terms · checked Sep 23, 2026
  5. [5]Anthropic, Pricing (API model table, the "Legacy models" heading, batch discount, fast mode, US-only inference). Vendor's own page · checked Sep 23, 2026
  6. [6]Anthropic, Commercial Terms of Service (training restriction, ownership of Inputs and Outputs, term and termination). Vendor's published terms · checked Sep 23, 2026
  7. [7]Anthropic, How long do you store my organization's data? (scopes itself to commercial products including the Anthropic API), Jul 1, 2026. Official documentation · checked Sep 23, 2026
  8. [8]Anthropic, Model deprecations (notice period, "not sooner than" dates, and which platforms the dates apply to). Official documentation · checked Sep 23, 2026
  9. [9]Anthropic, Claude API platform page (the clouds Claude is offered in). Vendor's own page · checked Sep 23, 2026
  10. [10]Microsoft, Data, privacy, and security for Foundry Models sold by Azure in Microsoft Foundry. Official documentation · checked Sep 23, 2026
  11. [11]vLLM project, vLLM documentation (the API surfaces the server exposes). Official documentation · checked Sep 23, 2026

Related

Free project intro call, 30 minutes.

We reply within 24 hours.

Book a Call