OpenAI has begun moving some enterprise customers to outcome-based pricing, charging only when an AI agent successfully completes a task rather than billing for raw compute. The undisclosed arrangement leaves the definition of success undefined, effectively turning AI evaluations from a quality check into billing infrastructure.
According to reports by The Information and The New Stack, OpenAI has started testing outcome-based pricing with some enterprise customers, billing on successful outcomes rather than on the compute or token counts consumed. The arrangement charges only when the AI successfully finishes a task such as handling a support request. The company had not previously disclosed such pricing.[1], [2]
Neither the product nor the customers covered by the arrangement were publicly named, and an OpenAI spokesperson declined to comment. The New Stack notes that OpenAI has not said how it determines when a task counts as a success, which turns AI evaluation tools from simple quality checks into core billing infrastructure.[1], [2]
No public document sets out the test for when a task counts as complete, leaving developers building agents to answer for themselves when a job is finished or how a disagreement would be settled. The shift marks a growing trend toward outcome-based pricing in the AI sector, drawing comparisons to Cognition, which offers credits worth up to 10 million dollars when its software fails to deliver.[1], [2]
