OpenAI licensing covers the terms under which OpenAI makes its models and applications available to organisations and individuals. For a software asset manager, the portfolio falls into three groups with different mechanics. The first is ChatGPT, sold per user per month in its paid plans, with credits or token-based billing on top for heavier use.[2] The second is the OpenAI API, billed per million tokens by model and processing mode.[3] The third is open-weight models and open source tools, such as the gpt-oss models and the Codex CLI, which are released under the Apache License 2.0.[15][17] OpenAI’s pricing FAQ states that the API is billed separately from ChatGPT Plus, Business, Enterprise and Edu.[2]
Business customers contract under the OpenAI Services Agreement, effective 2026-01-01, which governs the APIs, ChatGPT Enterprise, ChatGPT Business and other business services. Individuals use ChatGPT under the separate Terms of Use.[1][13] None of these products uses the per-core, per-processor or device metrics common in on-premises software. Compliance work centres on seat counts, consumption, credit balances, usage restrictions and data-use settings. This article summarises the position on 2026-10-01. Detailed prices and counting rules are in OpenAI ChatGPT and API licensing.
Product families
| Family | What is licensed | How it is licensed | Guide |
|---|---|---|---|
| ChatGPT Business | Workspace for teams; ChatGPT and Codex | Standard and Premium seats, monthly or annual, plus optional credits[4] | OpenAI ChatGPT and API licensing |
| ChatGPT Enterprise and Edu | Organisation-wide workspace; Edu is the university variant | Custom pricing; credit-based or token-based agreements[2][18] | OpenAI ChatGPT and API licensing |
| ChatGPT Go, Plus and Pro | Individual subscriptions | Per user per month under the Terms of Use[2][13] | OpenAI ChatGPT and API licensing |
| OpenAI API | Model access for developers and applications | Per 1M input, cached input, cache-write and output tokens, by model and processing mode[3] | OpenAI ChatGPT and API licensing |
| Scale Tier | Reserved API throughput for one model snapshot | Token units bought for a minimum of 30 days[12] | OpenAI ChatGPT and API licensing |
| gpt-oss models | Open-weight models gpt-oss-120b and gpt-oss-20b | Apache License 2.0 plus a short usage policy[15][16] | This article |
| Codex CLI | Open source coding agent client | Apache License 2.0; model use billed through ChatGPT or the API[17] | This article |
ChatGPT
The paid ChatGPT plans (Go, Plus, Business and Enterprise) “are priced per user per month”. Go, Plus and Business can be billed monthly, and Business and Enterprise annually.[2] ChatGPT Business is offered “For teams of 2–200 employees”. Its Standard seat costs $25 per user per month on a monthly plan, or $20 on an annual plan. Each Business seat type gives access to ChatGPT and Codex, and Premium seats have a higher usage allotment.[2][4] ChatGPT Enterprise has “Custom pricing”, and OpenAI offers credit-based and token-based pricing for it.[2] ChatGPT for Teachers is a free plan for verified U.S. K-12 educators through June 2027.[2]
OpenAI API
API prices are published “per 1M tokens” for each model, in columns for input, cached input, cache writes and output. The flagship table lists the processing modes Standard, Batch, Flex, Fast and Ultrafast, with separate short-context and long-context prices.[3] As of 2026-10-01 the flagship models are gpt-6-astra, gpt-6.1-sol and gpt-6-luna. At Standard short-context rates, gpt-6-astra costs $10.00 per 1M input tokens and $50.00 per 1M output tokens.[3]
Open-weight models and open source code
OpenAI releases the gpt-oss models under the Apache License, Version 2.0.[15] An accompanying usage policy states: “By using OpenAI gpt-oss-120b and gpt-oss-20b, you agree to comply with all applicable law.”[16] The Codex CLI repository is also licensed under the Apache License 2.0.[17] These components are handled like other open source software. Keep the license text and notices, and record them in the software inventory. See open source software licensing.
Metrics
The catalog records OpenAI’s metrics under OpenAI’s own names. The definitions are in the linked rows, and counting rules are in the guide.
| Metric | Where it applies | Role |
|---|---|---|
| Standard seat and Premium seat | ChatGPT Business | Purchased seats are billed “whether or not they’ve been assigned to individual members”; 2 seat minimum[4] |
| Codex seat | ChatGPT Enterprise (credit-based) | Codex-only access, usage-based[18] |
| Credits | ChatGPT Business, Enterprise and Edu | Flexible usage beyond seat limits; shared pools and quarterly true-ups on Enterprise[5] |
| Input tokens, cached input tokens, cache writes and output tokens | OpenAI API; token-based ChatGPT Enterprise | Price per 1M tokens by model and mode[3][6] |
| Token unit | Scale Tier | Tokens per minute for one model snapshot, minimum 30 days[12] |
OpenAI publishes no conversion between seats, credits and tokens. Rate limits on the API are set “at the organization level and at the project level, not user level”, so they do not map onto named users either.[11]
Purchasing programs and agreements
The Services Agreement is built around an Order Form, which is either a signed ordering document or the OpenAI webpage used to buy the Services. When buying through a reseller, the Order Form is the document signed with the reseller.[1] The main terms are:
- Term and renewal. The Services Term is listed on the Order Form. Seats added mid-term have “a prorated term ending concurrently with the then-current Services Term”. Notice of non-renewal or scope reduction must be given at least thirty days before the next Renewal Term. If the customer reduces its license count, quantity or minimum commitment, “OpenAI may adjust or remove discounts”.[1]
- Fees. Fees are payable to OpenAI or the customer’s reseller. They are non-refundable except as required by law, and a minimum commitment “is non-cancellable”. If the agreement ends other than by the customer’s termination for cause, unpaid minimum commitments become immediately due.[1]
- Price changes. “Price changes on the Pricing Page will be effective fourteen days after they are posted.”[1]
- Order of precedence. The Order Form prevails, then the Service-Specific Terms, then the Agreement, then the OpenAI Policies.[1]
- Credits. Prepaid and promotional credits are “non-transferable” and expire one year after purchase or issuance unless otherwise specified. Attempts to sell or transfer credits may lead to revocation or account suspension.[8]
The catalog’s program rows cover ChatGPT Business, ChatGPT Enterprise, ChatGPT Edu, ChatGPT for Teachers, OpenAI for Nonprofits, ChatGPT Plus and Pro and Scale Tier. OpenAI for Nonprofits offers “up to a 75% discount on ChatGPT Business or ChatGPT Enterprise”.[2]
Support and maintenance
OpenAI sells no separate maintenance product. Updates are delivered as part of the service. Under the Services Agreement, OpenAI warrants that the Services will conform “in all material respects with the Documentation”. If an update “materially reduces the Services functionality”, the customer may terminate on thirty days’ notice. That right must be used within five business days of OpenAI’s notice.[1] Beta Services “may not be supported and may be changed at any time without notice”. OpenAI has no liability for them.[1] The Enterprise plan lists “24/7 priority support, SLAs, custom legal terms” among its features.[2]
For the API, the equivalent of a product lifecycle is model deprecation. OpenAI notifies customers who use a model by email, and gives minimum notice before retirement “Unless safety or compliance concerns require a faster timeline”: at least 6 months for generally available models, and at least 3 months for specialised variants. Preview models may be retired with much shorter notice, such as 2 weeks.[14] Applications pinned to a model snapshot therefore need a migration plan in the same way as on-premises software needs an upgrade plan.
Audit and compliance
The Services Agreement contains no clause letting OpenAI audit a customer’s licence use in the on-premises sense. The audit clause runs the other way: on written request, at most once per year, OpenAI provides the customer with its most recent third-party security Audit Reports.[1] Compliance instead rests on metering, use restrictions and suspension rights:
- Metering. Usage-based fees are charged on usage that OpenAI calculates.[1] For a Business workspace, the billed quantity is the number of purchased seats, assigned or not.[4]
- Single-user accounts. End User Accounts may be used only by a single End User, and login credentials may not be shared.[1] For individual Pro plans, OpenAI’s help article restates prohibitions on sharing credentials and on “Reselling access or using ChatGPT to power third-party services.”[19]
- Use restrictions. Customers may not use Output to develop competing models except under a permitted exception, trade API keys, or circumvent rate limits.[1] The Usage policies apply across OpenAI products.[9]
- Suspension. OpenAI may limit or suspend access if required by law, if the customer violates the Agreement or OpenAI Policies, or to deal with a Security Emergency. It may also suspend or terminate when fees are past due.[1]
- Termination. Either party may terminate for material breach not cured within thirty days. On termination, OpenAI deletes Customer Content within thirty days unless legally required to keep it.[1]
Indemnity and output
OpenAI indemnifies business customers against third-party claims that the Services infringe intellectual property rights.[1] The Service Terms extend this, for API and Enterprise customers, to claims that the customer’s “use or distribution of Output infringes” a third party’s intellectual property right. Exclusions include cases where the customer knew or should have known the Output was infringing, or disabled OpenAI’s citation, filtering or safety features. Beta Services are excluded from indemnification.[7] The Services Agreement states that this Service-Specific Terms indemnity is not subject to the liability cap. OpenAI may not materially reduce it without the customer’s written agreement.[1] The Service Terms also warn that code generated by Codex “may be subject to third party licenses, including, without limitation, open source licenses”.[7]
Cloud, hosting and territorial rules
ChatGPT and the API are OpenAI-hosted services, so there is no bring-your-own-license question for them. The territorial rules are about where the services may be used and where data is processed:
- Supported countries. Customers and End Users “may not access or offer access to the Services outside of the Supported Countries and Territories”. OpenAI publishes separate country lists for the API and for ChatGPT.[1]
- Trade controls. The customer is responsible for sanctions and export compliance, and Input may not include material that requires a government licence for export.[1]
- Data residency. Regional processing endpoints are charged “a 10% uplift for models released on or after March 5, 2026”, and FedRAMP endpoints carry the same uplift.[3]
- Data use. OpenAI “will not use Customer Content to develop or improve the Services, unless Customer explicitly agrees to such use”.[1] The Enterprise privacy page applies the no-training default to ChatGPT Business, Enterprise, Edu and the API Platform.[10] Individual plans are different. Under the Terms of Use, OpenAI may use content to improve its services unless the user opts out.[13]
The gpt-oss models are the exception, because the licensee runs them. The Apache License 2.0 sets no limit on hardware, cloud or territory.[15]
Recent licensing changes
| Date | Change |
|---|---|
| 2026-01-01 | Current Services Agreement and Terms of Use took effect; Service Credit Terms updated[1][13][8] |
| 2026-03-05 | Regional processing endpoints charged a 10% uplift for models released on or after this date[3] |
| 2026-05-07 | Fine-tuning no longer available to organisations that had not previously run fine-tuning[14] |
| 2026-07-30 | Priority processing renamed Fast mode[3] |
| 2026-09-10 | New sign-ups and upgrades to ChatGPT Pro $200 paused[19] |
| 2026-09-29 | Service Terms updated[7] |
| 2027-01-06 | Announced: active existing customers can no longer create new fine-tuning jobs; inference on fine-tuned models continues until the base model is deprecated[14] |
Because the Pricing Page can change on fourteen days’ notice, published prices should be recorded with their retrieval date.[1]
Guides
| Article | What it covers |
|---|---|
| OpenAI ChatGPT and API licensing | ChatGPT plans and seat types, credits, token-based Enterprise billing, API token prices, Batch API, rate-limit tiers, Scale Tier, data use and retention, usage restrictions, programs |
| OpenAI Services Agreement and data controls | Document order and policy versions, workspaces and Affiliates, End User and account rules, competing-model restriction and Permitted Exception, ChatGPT retention and admin access, Zero Data Retention, Modified Abuse Monitoring, data residency, audit reports, DPA and exit |
Related articles: Anthropic licensing for a comparable vendor, generative AI licensing for the cross-vendor concepts and subscription and consumption licensing for the general model.
Out of scope
This article does not cover OpenAI models offered through third-party clouds under those providers’ terms, ChatGPT for Healthcare, Clinicians and Financial Services beyond the Service Terms notes above, Sora, image and voice prices, the Data Processing Addendum, or negotiated Enterprise and Scale Tier prices, which OpenAI does not publish.
Catalog rows
- Metrics: Standard seat, Premium seat, Codex seat, Credits, Input tokens, Output tokens, Token unit
- Programs: ChatGPT Business, ChatGPT Enterprise, Scale Tier
- Rules: ChatGPT Business 2 seat minimum, ChatGPT Business bills all purchased seats each cycle, One End User per End User Account, Usage-based fees calculated by OpenAI, No use of Customer Content to improve Services, Output and rate-limit restrictions