AI's Outcome Pricing Paradox

OpenAI president Greg Brockman said this at the launch of GPT-6 Astra in early September:

“Pricing tokens doesn’t make any sense… our tokens are not the same as our competitors’ tokens… what you actually want is the price per task.”

For the last few years, AI companies have told the rest of the industry to move from selling software as a subscription to selling outcomes: results as a service instead of SaaS.

The irony? Most of us pay a flat monthly AI subscription, paying the same amount whether the model helps us or not.

AI companies themselves have embraced the subscription model.

The companies pushing the shift to outcomes onto everyone else haven’t fully made it themselves yet.

Hopefully, we will see more business model experiments by AI companies.

To Brockman’s point, a more expensive model that solves a task correctly the first time can cost less overall than a cheaper model that gets it wrong and needs retries or a person to fix it.

OpenAI has started testing outcome-based pricing with some large enterprise customers, where they pay only when an AI agent successfully completes a task, and OpenAI absorbs the compute cost if it fails.

This is described as an early test with select large customers, not a broad rollout. OpenAI hasn’t disclosed exactly how “success” gets determined or what the pricing looks like.

References

• The Signal, 4 September 2026
• PYMNTS: OpenAI lets some customers pay only when AI performs
• The Information: OpenAI starts letting customers pay when AI works