Author
TL;DR: At its latest DevDay, OpenAI made a Plus or Pro subscriber’s included usage spendable inside partner apps. It also let eligible enterprises apply part of an existing OpenAI commitment to partner products built on OpenAI models. Both moves put a platform-funded path between an AI vendor and its money, whether a subscriber’s plan allowance or an enterprise commitment. Call that path the rail. For the vendor, three consequences follow:
- Budget. A broader committed budget that pre-empts deals before price comes up.
- Reference price. A public rate card that procurement will benchmark inference-shaped pricing against.
- Meter. A meter that moves from the vendor to the platform.
There are three defensible positions relative to the rail, set out below. Which one fits is a judgment about product shape and buying relationship inside your AI software pricing architecture.
- What OpenAI Changed at DevDay
- Why OpenAI’s Credit Design Is a Poor Template for Your Product
- Committed OpenAI Spend Becomes a Competing Budget Line
- What OpenAI Marketplace and Shared ChatGPT Credits Mean for AI Software Pricing: Distribution Versus Pricing Power
- Three Pricing Positions Relative to the Rail
- What to Do Before Your Next AI Pricing Change
- FAQs
What OpenAI Changed at DevDay
OpenAI named three separate programs. Its Marketplace help page separates the two enterprise programs and states that participation in one does not establish eligibility for the other. The consumer plan-usage program, which OpenAI calls a limited preview, is documented on its own page.
What is Sign in with ChatGPT plan usage?
Sign in with ChatGPT is OpenAI’s login for partner apps, and it now carries plan usage with it. A Plus or Pro subscriber’s included ChatGPT Work and Codex usage can be spent inside participating partner apps, under the rules in OpenAI’s plan usage help page. This is the program the phrase “shared ChatGPT credits” points at: what is shared is the plan allowance, and purchased credits enter only through the overflow below. Each app is capped weekly at a share of the plan’s usage, and the cap creates no separate balance.
An opt-in overflow to the subscriber’s own credits ships switched off, and it applies only when the app’s limit is set at the full allowance. OpenAI states that usage rates may differ between an app and ChatGPT. The partner’s own subscription and charges remain separate.
What does the OpenAI Marketplace change?
The OpenAI Marketplace changes the scope of an enterprise’s existing OpenAI commitment. An eligible customer can apply part of that commitment to a qualifying partner product built on OpenAI models. OpenAI and the partner confirm eligibility per customer, per product and per purchase, and OpenAI calls the program a beta.
The customer contracts with the partner and receives the partner’s invoice. OpenAI reconciles the eligible amount against the commitment. The Marketplace is a discovery page plus expressions of interest, with no self-service checkout. The published terms describe eligibility and reconciliation only, and say nothing about a margin, a referral fee or a resale role.
How does an organization’s OpenAI-funded inference run inside a partner app?
The third program lets an organization’s OpenAI-funded inference run inside a participating partner app. OpenAI’s Marketplace page treats this as a separate program and says that Marketplace participation does not establish eligibility for it. For the vendor, it is the enterprise shape of the meter question that plan usage raises: the inference is paid elsewhere, so the partner’s price has to attach to something else.
Why OpenAI’s Credit Design Is a Poor Template for Your Product
The plan-usage move made the consumer allowance portable, and it also made the allowance harder to read. One allowance now faces as many exchange rates as there are partner apps. OpenAI publishes its API and credit rates, but not the rate at which plan usage converts inside any given partner app. A subscriber can see how much allowance remains but cannot see what a given task inside a given app will cost.
The economics underneath differ by seller. OpenAI can price close to inference cost because it controls that cost curve. An application vendor reselling inference inherits the cost exposure without the cost control. The same unit therefore carries different economics for each party, a gap that hits hardest for vendors who control neither input cost nor consumption.
Expect “OpenAI does it” to surface in board meetings as a reason to adopt credits. Treating it as a pricing rationale imports a platform’s cost structure and market position into a business that has neither. That is a benchmarking error. The general case against inherited credit designs sits in our analysis of the structural flaws of credit-based pricing.
Committed OpenAI Spend Becomes a Competing Budget Line
Your buyer now holds a reference price for LLM inference
OpenAI publishes credit and token rate cards, and those cards give procurement a reference price for LLM inference. Any AI product whose pricing reads as “inference plus markup” invites a direct benchmark against them.
A value metric that counts units of work or business outcomes sits outside the rate card’s frame. That design removes the benchmark. We see the same pattern from the buyer’s side across engagements: a unit the buyer cannot forecast is negotiated off the contract and onto a structure the buyer can budget against.
Does committed OpenAI spend compete with your product for the same budget?
An enterprise commitment that can be drawn down on partner products is a broader budget than it was before DevDay. Expect a vendor off the rail to hear “we already committed that spend to OpenAI” before it hears a price objection. Lowering your price does nothing about that objection. The deal stalls on where the money sits, and the response is a packaging and positioning decision.
See how AI vendors are pricing around platform moves: Pricing Observatory
What OpenAI Marketplace and Shared ChatGPT Credits Mean for AI Software Pricing: Distribution Versus Pricing Power
Whose unit counts when the platform funds the purchase?
When inference is paid from the user’s plan, the vendor can no longer meter tokens it does not bill. The value metric has to move to the vendor’s own unit of work: runs, seats, outcomes. A vendor that priced on inference volume loses its meter the day a subscriber connects a ChatGPT plan.
What does joining the rail cost on the supply side?
Eligibility for commitment drawdown requires a product built on OpenAI models. Joining the rail therefore trades the freedom to source inference from more than one model vendor for distribution. It fixes the vendor’s cost side to one supplier whose prices move, and it removes the option to route workloads to a cheaper model when margins tighten.
Two channels, two renewal clocks
A vendor on the rail runs two channels, direct and platform-funded. Under Marketplace drawdown both invoice through the vendor, but they renew on different clocks: the vendor’s contract term and the customer’s OpenAI commitment cycle. That setup breeds discounting drift between channels, because each channel’s renewal pressure lands on price.
We have watched a sales team regularly upgrade customers to the richer edition and then discount it back to the previous edition’s price, because the customer needed only a few more dashboards. The dashboard cap sat on the edition boundary, so the only available increment was a whole-edition step. Reps closed the gap with price, and the reporting read a packaging signal as a discipline problem.
The rail changes where that signal lands. On the Marketplace the partner still sells, contracts and invoices, so a rep is in the loop and the same lever is on the desk. Now the customer’s committed OpenAI spend is the argument for using it. On the plan-usage rail there is no rep: the user signs in, and a boundary drawn in the wrong place shows up as a stalled upgrade rather than a discount. One renewal policy has to read both.
Three Pricing Positions Relative to the Rail
| Position | Who it fits | What the vendor keeps | What it gives up |
|---|---|---|---|
| 1. Complement the plan | Workflow products that sit beside ChatGPT and assume the customer already holds a plan | Its own unit of work as the value metric, and the direct customer relationship | Any claim on the inference spend, which the customer’s plan now carries |
| 2. Price outside the rail | Enterprise products with a direct, multi-stakeholder buying relationship | Full control of the value metric, the invoice and the renewal conversation | Access to the committed budget; the “already committed to OpenAI” objection has to be answered with packaging |
| 3. Join the rail | Self-serve add-ons whose users already hold a plan (plan usage), and products built on OpenAI models selling into committed accounts (Marketplace) | Reach into an installed base that already funds the purchase | The token meter, multi-model sourcing, and some control over price presentation and renewal timing |
In Position 1 the customer’s plan supplies the inference, so the reasoning in bring-your-own-model pricing applies directly. In Position 2 the vendor still decides whether to pass through inference costs or recast them into its value metric. In Position 3 the platform-funded offer works best as a deliberate entry edition, with the full pricing architecture kept in the direct channel.
Which position fits is a per-vendor judgment about product shape and buying relationship. A pricing expert can pressure-test that choice before a customer forces it.
Test the position you picked against market benchmarks: LevelSetter
Which of the Three Rail Positions Does Your Pricing Actually Occupy?
If your product assumes the customer already pays for ChatGPT, you’ve chosen a position relative to the rail, often by default. A few questions score your licensing, packaging, and pricing and show which decision that position puts at risk first.
What to Do Before Your Next AI Pricing Change
- Audit the unit. Does your value metric read as inference plus markup? If a buyer can divide your price by a token count, procurement will make that calculation.
- Map the budget. Where is your buyer’s AI budget already committed, and can that commitment now be drawn down on a competitor’s product?
- Pick a position. Which of the three positions fits your product shape and buying relationship, and does your sales team know the answer before a customer asks?
- Run the comparison. Read your pricing page the way procurement will, beside the published rate card. Which sentence invites the benchmark?
Find out whether your value metric survives a rate-card comparison: Pricing Architecture Assessment
The positions above are frames to test against your own product. If you are deciding whether to join the rail, price around it or stay off it, describe your product and the objection you are meeting, and a pricing expert will reply. Talk to an Expert.
