OpenAI Joins Salesforce in Pay-Per-Outcome AI Pricing Shift

What happened

OpenAI has quietly begun letting some of its largest customers pay only when its AI actually finishes a job, rather than billing them for access or for every token consumed along the way. The arrangement was first reported by Kevin McLaughlin and Amir Efrati at The Information, which cited a person familiar with the matter and said the option has been available for several months without any public announcement 9. PYMNTS picked up the report the same weekend, framing it as OpenAI giving "major customers" a new way to pay only when the company's AI "does its job" 114. TNW's write-up added color and skepticism in roughly equal measure, noting it could not independently verify the arrangement and that the customers, prices and exact contract terms are still unknown 10.

The example cited across the coverage is customer support: instead of paying for API calls or software access, a customer pays when an AI system resolves a support interaction end-to-end 91013. OpenAI has not disclosed which customers are involved, what they're being charged, or how success is defined, and the company declined to comment through intermediary reporting 13.

Why this matters beyond OpenAI

The move slots into a broader pivot the industry is calling outcome-based pricing, in which vendors bill for a completed result — a resolved ticket, a qualified lead, a closed sale — instead of a subscription seat or a metered unit of computing. Salesforce is the most-cited parallel. CEO Marc Benioff told investors last week that "customers want to buy and want to price in different ways," and the company now lets enterprise clients choose among per-seat licensing, consumption credits, or contracts tied to revenue gains or cost savings attributed to its Agentforce AI 11415. Salesforce has gone further with a specific product, Agentforce Help Agent, charging $2 per successful autonomous resolution, with no fee if the case escalates to a human or receives negative customer feedback 1617. That $2 figure sits alongside — not in place of — Salesforce's existing per-user licensing and Flex Credit options, which price individual AI actions at roughly ten cents apiece 17.

Intercom offers the clearest public benchmark for how this pricing already works in production. Its Fin agent charges $0.99 per resolution and nothing for a failed or escalated conversation, with a resolution defined narrowly as either explicit customer confirmation or the customer not asking for further help 1819. Intercom's newer sales product prices a qualified lead at $9.99, more than ten times the fee for an ordinary support resolution, illustrating how outcome pricing can be calibrated to the underlying value created rather than the effort expended 2010.

Why vendors are trying this now

The reporting converges on a shared explanation: AI products are expensive to operate, IT budgets are strained, and software companies have not yet seen AI translate cleanly into faster revenue growth 11314. Token-metered billing has made that tension acute — TNW cites one developer who ran a hundred agents in parallel and racked up $1.3 million in OpenAI token charges in thirty days, an extreme illustration of costs scaling with attempts rather than results 10. Research from Futurum Group, referenced in that same coverage, found 43% of enterprise buyers now prefer consumption-based pricing and 27% prefer outcome-based pricing, with fewer than one in five still wanting to pay strictly per seat 10. OpenAI's own published API pricing remains firmly consumption-based for now — a sprawling table of per-token rates spanning from $0.05 to $30 per million input tokens depending on the model — which underscores that the outcome-based option reported by The Information is a selective carve-out for certain large accounts, not a replacement for OpenAI's public price list 1112.

The hard part: defining success

Every outlet that goes beyond the headline converges on the same complication — deciding what counts as a completed task, and who gets credit for it. PYMNTS highlights Stripe's guidance on outcome-based pricing, which warns that a sales conversion or cost saving "could stem from product tweaks, marketing campaigns or seasonality" rather than the software itself, and that without explicit attribution rules, customers and vendors can end up disputing whose success it was 114. Superpower Daily frames this as the central commercial question for OpenAI: whether a completed task can become a "shared, defensible measure of value" rather than just a technical demonstration 13. Constellation Research, covering Salesforce's rollout, makes a similar point about "haggling over definitions and fuzzy math" even when a vendor tries to build in safeguards, such as not charging when a customer gives negative feedback or requests a human agent 16.

Where the reporting agrees

Across the outlets, several core facts hold steady without contradiction. All confirm that The Information broke the OpenAI story on August 30, 2026, based on a source familiar with the arrangement, and that OpenAI has not made any public announcement 19101314. All agree the practice is limited to select large accounts, not a general pricing change 91013. All point to customer support as the illustrative use case, with resolution of a support interaction serving as the model example of a definable, billable outcome 91014. And every outlet that discusses the wider market — PYMNTS, TNW, Superpower Daily, and the Salesforce-focused pieces — agrees that OpenAI's move sits inside a larger industry pattern already visible at Salesforce, Intercom, Sierra, Cognition and Adobe, all of which have introduced some form of outcome- or usage-linked pricing for AI products 110131617. There is also consistent agreement that this does not eliminate subscription or per-seat pricing; instead, outcome pricing is being layered on top of, or offered alongside, existing licensing and consumption models at both OpenAI and Salesforce 161719.

Where it doesn't

The coverage diverges mainly on framing and specificity rather than on hard facts, and it's worth being precise about which claims are attributed versus asserted. PYMNTS treats the shift as part of a "structural recalibration" of SaaS economics — a permanent restructuring of how software is priced now that AI agents do work previously requiring human staff 114. TNW is more cautious, repeatedly stressing that it has not independently verified The Information's account and that OpenAI's customers, prices and terms remain entirely unknown, treating the story as a significant but still-thin data point 10. Superpower Daily leans hardest into the unresolved attribution problem, treating it almost as the real story rather than a footnote 13.

There's also a difference in how concretely each outlet can describe outcome pricing in practice. Salesforce and Intercom have public, numbered prices — $2 per resolution for Agentforce Help Agent, $0.99 per resolution and $9.99 per qualified lead for Intercom's Fin 16171920. OpenAI's arrangement has no published price at all; every account of it relies on a single unnamed source cited by The Information, repeated without independent confirmation by PYMNTS and TNW 191014. That's a meaningful gap: the Salesforce and Intercom pricing is verifiable and public, while the OpenAI story is, so far, one outlet's sourcing echoed by others.

The reading the evidence supports

Taken together, the sources support treating this as a real but narrow development rather than evidence that OpenAI has overhauled its pricing. The underlying claim — that OpenAI is testing pay-on-completion deals with select customers — rests on a single original source, and no outlet has added independent confirmation, pricing detail, or customer names beyond what The Information reported. What the evidence solidly establishes is the surrounding trend: Salesforce, Intercom, Sierra, Cognition and others have already built public, priced, outcome-based products, which makes OpenAI's quiet experiment plausible and consistent with where enterprise AI pricing is visibly heading, even though the OpenAI specifics themselves remain unverified. The more defensible story, then, is not that OpenAI has reinvented software billing, but that it is following a pattern others have already made public — and doing so cautiously, without the pricing transparency its peers have adopted.