OpenAI pilots pay‑per‑result pricing

OpenAI has begun testing outcome-based pricing with a small group of enterprise customers, according to multiple reports published on August 31 and September 2, 2026. Under the pilot, customers are billed when an AI system completes a defined task rather than for raw token consumption or a flat subscription, the reports said. OpenAI has not announced the program on its own website, and its public business and API pricing pages still center on seats, credits and token-based charges. ### What exactly is OpenAI testing? Multiple reports said the pilot applies to select enterprise accounts and ties charges to a measurable result, such as a task being successfully completed. The public reporting did not identify participating customers, covered workflows, contract length or the exact definitions OpenAI is using to judge success. (ai-market-watch.com) OpenAI’s own pricing pages, by contrast, continue to present standard commercial models. The business pricing page lists per-seat plans, while the API pricing documentation lists model charges by input, cached input and output tokens; the Help Center also references enterprise rate cards and credit-based billing. That gap suggests the pay-per-result model is, for now, a negotiated pilot rather than a published default. ### Why does this matter if token pricing still exists? (ai-market-watch.com) Token pricing charges for computation whether a workflow succeeds or fails. Outcome pricing shifts part of that risk to the vendor by linking payment to a pre-agreed result. In practice, that can make enterprise buyers focus less on model usage and more on whether a system resolved a support case, completed a document workflow or finished another business process. (openai.com) The reporting around OpenAI’s pilot also points to a narrower commercial question: how to prove that the AI, rather than a human fallback or outside factor, caused the result. One report cited Stripe guidance warning that a “successful” outcome can reflect changes in marketing, product design or seasonality, not just software performance. ### Where would this kind of pricing work first? (ai-market-watch.com) Customer support and tightly scoped back-office workflows are the clearest early candidates because they already produce measurable endpoints. A contract can define whether a ticket was resolved, a lead was processed or a document was completed, and both vendor and customer can audit the result. That makes outcome billing easier than in open-ended creative or research tasks. This is an inference based on the types of examples cited in the reporting and comparator pricing pages. (completeaitraining.com) Salesforce’s public Agentforce pricing already offers consumption-based options and, in some products, pay-per-resolution language. Adobe told investors in a 2026 summit presentation that its business models are expanding to include consumption and outcome-based pricing alongside subscriptions. Those disclosures show that large software vendors are at least publicly framing AI pricing around business results, not just software access. (thevalue.engineering) ### What changes inside the product when billing depends on results? Success measurement becomes part of the product, not just the finance system. A vendor needs a definition of completion, logs showing whether the system met that definition, and rules for edge cases such as partial completion, human intervention or retries. Without that, billing disputes become product disputes. This is an inference drawn from how outcome-based contracts are described in the reporting and comparator materials. (salesforce.com) Failure handling also becomes commercial infrastructure. If a workflow falls back to a human, stalls midway or produces an answer that must be corrected, the contract has to say whether that counts as success, partial success or no charge. Those details were not disclosed in the OpenAI reports, but they are central to any pay-for-performance model. (ai-market-watch.com) ### Is OpenAI replacing its existing pricing? OpenAI’s public materials do not show a broad replacement of token or seat pricing. The company’s business, API and Help Center pages still describe conventional pricing structures, including token-based API rates and enterprise credit or rate-card arrangements. That indicates the pilot is additive, at least for now. The next concrete sign to watch is a change on OpenAI’s official pricing or enterprise pages, or a named customer case study describing an outcome-based contract. (ai-market-watch.com) Until then, the evidence supports a limited enterprise pilot reported by outside outlets, not a public platform-wide pricing change. (openai.com)