OpenAI: the pricing arc
OpenAI's tracked moves stay close to the rate card. A reasoning model was cut steeply on the day a premium variant launched above it, setting the pattern of a falling floor with a new rung over it. The flagship family then shipped as three named rungs in one announcement, and the rungs moved separately soon after, the smallest one far more than the top. Promotional pricing arrived with a published end date, so the reversion is an increase that needs no announcement. The one step away from tokens is on the enterprise plans, where the pricing page discloses credit-based billing beside token-based billing. Everywhere else on this ledger the unit is still the token.
tracked and read.
first move to latest.
last re-checked.
Last tracked move:
OpenAI pricing moves, newest first.
- Jun 10, 2025Pricing · Rate cut
- Apr 24, 2026Pricing · Rate card
- Jun 26, 2026Packaging + Pricing · Rate card
- Jul 09, 2026Pricing · Rate cut
- Jul 30, 2026Pricing · Rate cut
- Aug 06, 2026Packaging + Pricing · Allowance up
- Aug 09, 2026Licensing + Packaging · Unit swap
- Aug 10, 2026Packaging + Pricing · Rate card
- Aug 13, 2026Licensing + Packaging · Repackage
- Aug 21, 2026Pricing · Rate cut
- Aug 30, 2026Packaging + Pricing · Repackage
- Sep 10, 2026Pricing · Rate card
- Sep 29, 2026Pricing · Price hold
- Sep 29, 2026Licensing + Pricing · Allowance cut
- Sep 29, 2026
- Sep 29, 2026Packaging + Pricing · Repackage
- Sep 29, 2026Terms change
- Sep 29, 2026Licensing · Terms change
- Sep 29, 2026Licensing + Packaging · Unit swap
- Oct 06, 2026Licensing + Packaging · Access change
The move
On October 6, 2026 OpenAI replaced its five numbered API usage tiers with three named tiers, Build, Launch and Grow. Build is reached at $5 in cumulative credit purchases with a $500 monthly spend cap, Launch at $100 with $5,000, and Grow at $500 with $200,000; before the change the ladder ran from Tier 1 ($5 paid, $100 a month) through Tier 5 ($1,000 paid, $200,000 a month), so the top cap now opens at half the prior spend and the lower caps rose. Concurrent GPT-Live voice sessions follow the new tiers at 50, 300 and 500.
The SPP read
OpenAI replaced the numbered usage tiers with a shorter ladder of named tiers. These tiers set how much an API account may spend on the token meter each month. The top monthly spend cap now opens at half the cumulative credit purchase it previously required, and the caps on the lower rungs rose. No per-token rate moved: what changed is how much metered spend an account may run at each level of prior purchase, and GPT-Live voice concurrency now follows the same ladder.
The move
OpenAI's 'Sign in with ChatGPT' lets Plus and Pro subscribers use their ChatGPT plan for AI requests in participating third-party apps and sites, with eligible requests counting toward the ChatGPT Work and Codex usage included in their plan, without needing to create or share an OpenAI API key.
The SPP read
Participating apps and sites can now run eligible AI requests against a subscriber's ChatGPT Plus or Pro plan. Those requests draw down the Work and Codex usage already included in the subscription, at rates the vendor says may differ between an app and ChatGPT, under a weekly cap per app set as a share of the plan's usage. The plan fee, rather than a separate API meter, covers inference done inside another vendor's product, and the partner's own subscription and charges stay separate.
The move
When users reach their ChatGPT plan limits, third-party apps connected via Sign in with ChatGPT can draw on available ChatGPT credits, but only if the user has explicitly opted in. This setting is off by default.
The SPP read
Once a subscriber exhausts the plan's included usage, apps connected through Sign in with ChatGPT can draw on that subscriber's available ChatGPT credits. The setting ships off, applies across participating apps at once, and works only for an app whose weekly limit is set to the full plan allowance. That makes a personal credit balance an overflow meter for third-party products, with the brake held by the user.
The move
OpenAI launched OpenAI Marketplace, a program that lets eligible enterprise customers apply part of their OpenAI spending commitment toward qualifying partner products built on OpenAI models, rolling out alongside DevDay announcements.
The SPP read
The move is in the terms around OpenAI's enterprise spending commitment, not in any published rate: part of a commitment can now be applied to qualifying third-party products built on OpenAI models. This follows the cloud-marketplace drawdown pattern and widens what prepaid OpenAI spend can buy, with no rate movement stated. Only eligible enterprise customers and qualifying partner products are in scope.
The move
OpenAI made its Ultrafast service tier generally available for GPT-6 Astra in the Responses API, selectable per request by setting service_tier to 'ultrafast' and priced as its own tier on the API pricing page, following the August limited preview on GPT-5.6 Sol that carried no rate.
The SPP read
OpenAI's Ultrafast lane moved from a limited preview with no rate attached to general availability on GPT-6 Astra with its own published price, selected per request through the service_tier parameter. The unit stays tokens; what moved is the packaging, with latency now sold as a priced option on the same model. The register's August row holds the preview; this row is the rate arriving.
Read revised Oct 20, 2026Reason: Chris 10-20 review: the row is the rate arriving after the preview (596); duplication check
OpenAI added a faster service tier for GPT-6 Astra in the Responses API. Developers select it on each request through the service_tier parameter instead of calling a different model. The unit stays tokens; what moved is the packaging, with latency now sold as its own priced option on the same model.
Claim revised Oct 20, 2026Reason: Chris 10-20 review: state the act (GA with a price) and its relation to the August preview row 596 instead of describing the docs
OpenAI's API documentation says developers enable the Ultrafast tier per request by setting service_tier to 'ultrafast', making speed a selectable priced tier rather than a separate model.
The move
OpenAI launched ChatGPT Pro 500, a new $500/month plan that includes Astra Ultrafast and the highest included usage of its Pro plans.
The SPP read
OpenAI added a plan above its existing Pro subscription, with the highest included usage in the Pro lineup and the only access to the Ultrafast mode in ChatGPT Work and Codex. The unit is still a flat subscription with an included allowance; what moved is the top of the ladder, where speed and allowance now sell as an edition of their own, and buying credits on a lower plan does not reach them.
Read revised Oct 21, 2026Reason: Chris 10-21 'flatten them': the contrast clause removed, facts unchanged
OpenAI added a plan above its existing Pro subscription, with the highest included usage in the Pro lineup and the only access to the Ultrafast mode in ChatGPT Work and Codex. The unit is still a flat subscription with an included allowance; what moved is the top of the ladder, where speed and allowance now sell as an edition of their own instead of as extra credits bought on a lower plan.
The move
OpenAI reopened new ChatGPT Pro 200 subscriptions at an unchanged $200/month price but with a lower usage allowance than previously offered, which it attributed to its increasingly efficient models.
The SPP read
New ChatGPT Pro buyers receive less included usage on the flat monthly subscription, while the list price stays the same, so the effective cost per unit of work rises with no rate moving. Subscribers eligible for grandfathering keep the earlier allowance. OpenAI describes the reduction as a consequence of model efficiency.
The move
OpenAI says GPT-6.1 Sol nearly matches GPT-6 Astra's intelligence on agentic coding, computer use, and professional work at one-fifth of Astra's standard input and output token prices.
The SPP read
OpenAI released GPT-6.1 Sol, a point update to its lower-priced Sol tier, and pitched it as close to GPT-6 Astra on agentic coding, computer use and professional work. The one-fifth figure compares two tiers' standard token rates; no existing rate moved. The Sol tier kept its prior token pricing while the capability on it moved up, which makes this a hold on per-token rates.
Read revised Oct 20, 2026Reason: desk pass 10-19: removed the X-not-Y and rather-than shapes and the meta line; facts unchanged
OpenAI released GPT-6.1 Sol, a point update to its lower-priced Sol tier, and pitched it as close to GPT-6 Astra on agentic coding, computer use and professional work.
The move
OpenAI bills GPT-Live 1 voice sessions at $0.05 per minute, metered per second without rounding up to a whole minute, with backend model and tool usage charged separately. Concurrent voice sessions are capped by the organization's usage tier, from 25 sessions on the lowest paid tier to 500 on the highest at launch, with tiers reached by cumulative credit purchases.
The SPP read
With GPT-Live 1 reaching general availability in the API, OpenAI has put a published rate on voice sessions charged by elapsed time and metered per second, with no rounding up to whole minutes. That session meter covers only the voice layer; backend model and tool usage is charged separately, so a session's total cost stacks a time-based meter on top of OpenAI's other usage charges. Capacity is packaged as well: the number of voice sessions an organization may run at once is set by its usage tier, and tiers are reached by cumulative credit purchases, so the session ceiling rises with what the customer has already spent.
Claim revised Oct 21, 2026Reason: verifier correction applied at the desk 10-21: at the 09-10 launch the caps ran by numbered usage tiers; the Build/Launch/Grow names arrived 10-06
OpenAI bills GPT-Live 1 voice sessions at $0.05 per minute, metered per second without rounding up to a whole minute, with backend model and tool usage charged separately. Concurrent voice sessions are capped by the organization's usage tier: 50 on Build, 300 on Launch, 500 on Grow, with tiers reached by cumulative credit purchases.
Read revised Oct 21, 2026Reason: Chris 10-21 round 3: capacity as a packaged capability by usage tier
With GPT-Live 1 reaching general availability in the API, OpenAI has put a published rate on voice sessions charged by elapsed time and metered per second, with no rounding up to whole minutes.
The move
OpenAI restructured the ChatGPT Business plan by introducing two distinct seat tiers—a Standard seat at $20/month and a Premium seat at $100/month (with monthly billing at $25 and $125 respectively)—replacing the previous single $20/user/month pricing, while also updating included features and removing the 2+ user minimum in favor of a 2–200 employee range.
The SPP read
On August 30, 2026, OpenAI's ChatGPT Business plan offered two kinds of seat where it had offered one: Standard seats at the existing per-seat rate and Premium seats at a higher one, with the plan's user range widened from a minimum to a stated band of company sizes. The seat stayed the unit; what changed is that one workspace now buys two kinds of it, priced apart, and the Premium rate arrived alongside a launch credit that has since ended.
Read revised Sep 15, 2026Reason: Read revised to describe the change without repeating the price points, which stay in the move record.
On August 30, 2026, OpenAI's ChatGPT Business plan carried two seat tiers where it had carried one: Standard seats at the existing per-seat rate and Premium seats at a higher one, with the plan's user range widened to two hundred seats.
The move
OpenAI cut the prices of its frontier GPT-5.6 Sol model for developers by more than 20% for the next three months, effective August 21, 2026.
The SPP read
What moved is the developer API rate on OpenAI's current frontier model; the token meter and how usage is counted stay the same. The cut is time-boxed, so it works as a promotional discount against the standing rate card. It keeps OpenAI's frontier tier on the tokens end of the arc and puts pressure on competing frontier rates while the window is open.
Read revised Oct 20, 2026Reason: desk pass 10-19: removed the X-not-Y and rather-than shapes and the meta line; facts unchanged
What moved is the developer API rate on OpenAI's current frontier model; the token meter and how usage is counted stay the same. The cut is time-boxed rather than a new list price, so it works as a promotional discount against the standing rate card. It keeps OpenAI's frontier tier on the tokens end of the arc and puts pressure on competing frontier rates while the window is open.
The move
On August 13, 2026, OpenAI opened a limited preview of Ultrafast, a new service tier that runs GPT-5.6 Sol up to 14 times faster than Standard processing, to a select group of API customers, with access to expand as capacity grows and no rate published.
The SPP read
On August 13, 2026, OpenAI opened a limited preview of Ultrafast, a separate service lane that runs GPT-5.6 Sol at a multiple of Standard processing speed, to a select group of API customers, with no rate published. The lane is defined by speed on the same model, so the pricebook now holds a slot with a service level and no price. The rate, when it prints, will show what a faster lane on an unchanged model is priced at.
Read revised Sep 15, 2026Reason: Read revised to describe the change without repeating the price points, which stay in the move record.
On August 13, 2026, OpenAI opened a limited preview of Ultrafast, a service tier that runs GPT-5.6 Sol up to 14 times faster than Standard processing, to a select group of API customers, with no rate published.
The move
On August 10, 2026, OpenAI announced Premium seats for ChatGPT Business at $125 per month or $100 per month billed annually, alongside the standard seats at $25 monthly or $20 annually, with $100 in workspace credits per qualifying Premium seat for the first 10,000 waitlist workspaces, up to five seats; the credit offer closed by late August.
The SPP read
On August 10, 2026, OpenAI announced Premium seats for ChatGPT Business at a higher per-seat rate than the standard seats, monthly or annual, and attached a workspace credit per Premium seat for an early cohort of waitlist workspaces, capped at a handful of seats each. The credit closed by late August; the seat class stayed. Read with the Business restructure that followed on August 30, the arc shows a seat unit splitting into two priced kinds inside one plan, the standard seat price unchanged.
Read revised Sep 15, 2026Reason: Read revised to describe the change without repeating the price points, which stay in the move record.
On August 10, 2026, OpenAI announced Premium seats for ChatGPT Business at a higher per-seat rate than the standard seats, monthly or annual, and attached a workspace credit per Premium seat for the first ten thousand waitlist workspaces, capped at five seats each.
The move
OpenAI added a footnote to Enterprise pricing clarifying that both credit-based and token-based pricing are available for Enterprise plans, and relabeled the 'Messages and interactions' feature row to 'Everyday text chats' while removing GPT-5.5 Instant as a listed model from Business and Enterprise plans.
The SPP read
OpenAI updated its Enterprise pricing page to formally disclose that both token-based and credit-based billing are available on Enterprise plans, marking a structural clarification of which units buyers may be metered on at that tier. Separately, a listed model was removed from the Business and Enterprise plan feature matrix, narrowing the visible access set for those tiers. A feature row was also relabeled, repositioning how the capability is surfaced to buyers without a stated change to underlying limits.
The move
On August 6, 2026 OpenAI announced that GPT-5.6 Luna would replace GPT-5.5 Instant as the default model for ChatGPT Free and Go users that week, with unlimited everyday text chats (limits still apply to file uploads, images and other tools) and a new Think button for harder questions arriving the following week.
The SPP read
OpenAI's post of August 6 set a two-step change for the Free and Go tiers: a new default model that week, then unlimited everyday text chats and a Think button the week after, with limits kept on uploads, images and other tools. Once the second step landed, message volume on the bottom tiers no longer separated them from paid plans; model access and reasoning depth became the remaining differences. On OpenAI's arc this is an allowance change at a held price of zero.
Read revised Oct 21, 2026Reason: Chris 10-21 round-4 validation: the two-step rollout stated as OpenAI's post states it
The included chat allowance on ChatGPT's free and Go tiers changed from capped to unlimited for everyday text chats on GPT-5.6 Luna.
Claim revised Oct 21, 2026Reason: verifier FACT_OFF correction applied at the desk (Chris 10-21 round 4): the rollout was staged, default model that week, unlimited chats and Think the week after
On August 6, 2026 OpenAI gave ChatGPT Free and Go users unlimited everyday text chats on GPT-5.6 Luna, which became the default model for those tiers, and added a Think button for harder questions.
The move
On July 30, 2026, OpenAI cut GPT-5.6 Luna prices by 80%, to $0.20 per million input tokens and $1.20 per million output tokens, cut GPT-5.6 Terra prices by 20%, to $2 per million input tokens and $12 per million output tokens, and added a Fast mode for GPT-5.6 Sol, per OpenAI's official announcement.
The SPP read
One announcement, three prices. On July 30, 2026 OpenAI cut GPT-5.6 Luna steeply, trimmed GPT-5.6 Terra by a smaller step, and added a Fast mode for Sol. The headline write-ups reported Luna's cut; the card listed all three. The cheap rung got radically cheaper while the flagship edged down, which is what a price-performance frontier looks like when a vendor moves the whole curve at once.
Read revised Sep 15, 2026Reason: Read revised to describe the change without repeating the price points, which stay in the move record.
One announcement, three prices.
The move
GPT-5.6 (Sol, Terra, Luna) launched on July 9, 2026 and became OpenAI's new flagship family, with Sol matching GPT-5.5's $5/$30 pricing; GPT-5.5 remains available but is no longer the current flagship.
The SPP read
GPT-5.6 launched July 9 with Sol as the flagship at the same list price as the GPT-5.5 it superseded. The August 22 move makes this an arc event: Sol's price went on a clock, with the developer pricing page listing a lower promotional rate that holds at least through November 21, 2026. A discount with a published end date is a promotion, and its withdrawal is an increase nobody has to announce. Google shipped the same instrument in the same window, with Gemini prices that revert on January 1, 2027.
Read revised Sep 15, 2026Reason: Read revised to describe the change without repeating the price points, which stay in the move record.
GPT-5.6 launched July 9 with Sol as the new flagship at $5 per million input and $30 output on short context, the same list price as the GPT-5.5 it superseded; OpenAI positioned Terra, at half that, as the 5.5 analog.
The move
OpenAI stated that GPT-5.6 Luna is priced at $1 per 1M input tokens and $6 per 1M output tokens.
The SPP read
One announcement, three named rungs: a small tier at a fraction of the mid, a mid at half the top, and the top holding the flagship rate. Model families used to ship as one price with a mini beside it; this ships as a designed ladder, and the design is the point. Named rungs give the meter a packaging layer: buyers self-sort by workload instead of negotiating, and the vendor gains rungs it can move independently later. The ledger shows that option getting exercised within weeks, when the bottom rung was cut by four fifths and the middle trimmed while the top held. A ladder is the set of places a future price change can land without touching the others, and this vendor built one on purpose.
The move
OpenAI charges $5 per million input tokens and $30 per million output tokens for API use of GPT-5.5.
The SPP read
A six-to-one output premium puts the price on generation, not context. For application vendors pricing on top of the API, the asymmetry matters more than the headline rate, because features that read long and write short have a different unit cost than features that write long. Blended per-request pricing built on an average mix breaks the day the mix shifts. Price the direction your workload actually spends.
The move
OpenAI cut the price of o3 by 80%, with new pricing of $2 per 1M input tokens and $8 per 1M output tokens, effective June 10, 2025.
The SPP read
A deep cut on the reasoning model, announced the same day a pro variant arrived at a multiple of the new rate, is one decision expressed as two prices. The cut moves the base model toward commodity territory and the simultaneous premium launch keeps a rung above it for the spend to reach. The price of last year's frontier falls, and a new rung appears above it, so the falling floor never brings the ceiling down with it. Read the gap between the rungs rather than either price alone.
Read revised Sep 15, 2026Reason: Read revised to describe the change without repeating the price points, which stay in the move record.
An 80% cut on the reasoning model, announced the same day a pro variant arrived at ten times the new rate, is one decision expressed as two prices.
Planning a move like one on this arc, or countering one OpenAI just made? Talk to a pricing expert before it ships: describe the move, and a pricing architect reads it against your own licensing, packaging, and pricing.
Is your own pricing architecture ready for a move like this?
Every OpenAI move on this arc changed its licensing, packaging, or pricing. Score your own architecture, twenty-five questions in four minutes, free, and see which of the three holds the risk.
What we have written about OpenAI.
When Your Credit Spends in Someone Else’s Product: What OpenAI’s Plan Sharing and Marketplace Mean for AI Software Pricing
OpenAI Marketplace lets enterprise commitments fund partner products, and plan sharing spends ChatGPT usage in partner apps. What both change for AI pricing.
Inner and Outer Agent Harness Pricing Explained
The inner harness handles model invocation. The outer harness decides who pays for it. See why pricing belongs at the outer layer.
Where Your Software May Run Is a Licensing Decision, Not a Deal Term
Deployment is one of four dimensions in a software license grant. Where the software runs, and where the model runs, is priced before it is negotiated.
Other vendors on the log.
All tracked vendors on the AI Pricing Observatory
Get this week’s verified AI pricing moves in your inbox.
The moves, the week they’re verified. A dated headline and a one-line read for each new move, with a link back to the full page.
One issue a week, only when something’s actually moved. Unsubscribe anytime from the confirmation email.
Prefer Slack? Get the weekly digest in Slack.
Primary sources behind the OpenAI arc, dated.
| Date | Source | Record |
|---|---|---|
| developers.openai.com | primary ↗ | |
| help.openai.com | primary ↗ | |
| help.openai.com | primary ↗ | |
| help.openai.com | primary ↗ | |
| developers.openai.com | primary ↗ | |
| openai.com | primary ↗ | |
| help.openai.com | primary ↗ | |
| openai.com | primary ↗ | |
| developers.openai.com | primary ↗ | |
| openai.com | primary ↗ | |
| community.openai.com | primary ↗ | |
| openai.com | primary ↗ | |
| openai.com | primary ↗ | |
| openai.com | primary ↗ · archived | |
| openai.com | primary ↗ | |
| openai.com | primary ↗ · archived | |
| openai.com | primary ↗ · archived | |
| openai.com | primary ↗ · archived | |
| openai.com | primary ↗ · archived | |
| 2025 | ||
| community.openai.com | primary ↗ · archived | |
Chris Mele
Ranked #1 on OpenView’s list of B2B SaaS pricing experts. Every read on this page is written by a pricing architect, never drafted by a model, backed by a team that has held CFO, CPO, and CIO seats inside software companies.
The claims ledger and citability gate do the heavy lifting on sourcing. The expert makes the calls only humans can: what a move actually changes under its label, and what it signals for the vendors around it.
Read more about Chris →Want this arc read against your own pricing architecture?
Watching how OpenAI moves is one input. Deciding how your own licensing, packaging, and pricing should respond is a different discipline, and it is not something a dashboard can do for you.