OpenAI has quietly started letting some large customers pay only when its AI actually completes a task, according to The Information. The arrangement, which hasn’t been reported before, lets companies like customer support operations pay upon successful resolution rather than a flat subscription. An OpenAI spokesperson declined to comment.
The shift puts OpenAI in line with a pricing model that startups have been pushing hardest. Sierra and Fin — the latter being acquired by Salesforce for $3.6 billion — only charge when their AI finishes work without human intervention. Coding assistant Cognition goes even further, promising corporate clients up to $10 million in credits if the software fails to deliver value at least equal to what they paid. Adobe, HubSpot, and Zendesk are moving in the same direction.
The pressure comes from both sides. AI products are expensive to run, and software companies haven’t seen the revenue acceleration they hoped for. Meanwhile, IT budgets are stretched thin by competing tools like Anthropic’s Claude, which recently leaned harder into usage-based billing. Salesforce now lets customers negotiate Agentforce contracts tied directly to revenue lifts or cost reductions. CEO Marc Benioff framed it bluntly: the goal isn’t charging two dollars because the software closed some calls — it’s charging two dollars because the AI brought in 20 or 40.
The thorniest problem is attribution. Stripe has already published guidelines warning that a result could stem from product changes, marketing, or seasonality rather than the AI itself. Without clear rules for who gets credit, customers can dispute whether the vendor actually drove the success. Adobe is tiptoeing in too, billing part of its new CX Enterprise suite by value created — like ad campaigns completed — though it hasn’t named prices. The subscription model isn’t dead, but it’s suddenly negotiable in ways that would have seemed radical two years ago.
