OpenAI has begun offering some major enterprise customers the option to pay only when its AI successfully completes a task, moving away from the standard token-based billing model that has defined the industry1,2. Salesforce and other AI providers are also testing outcome-based pricing.
The shift, first reported by The Information, means OpenAI is billing select enterprise customers based on successful outcomes rather than the volume of compute consumed. Under the conventional token-pricing model, an AI agent can burn through tokens retrying failed steps without ever finishing a job, and the customer still pays. Under outcome pricing, failed runs become the provider's expense.
ANALYSIS The model inverts the financial risk: OpenAI, not the customer, absorbs the cost of unsuccessful attempts.
The approach raises a concrete technical question: how to determine when a task counts as successfully completed. Some outcomes are straightforward for software to verify, such as a support ticket that closes without human involvement. Others are ambiguous. A coding agent might rewrite code and pass every test, only for the patch to cause a new problem in production; the agent technically completed the task, but the customer would likely not consider it a success.
OpenAI's hosted tooling can check responses against expected results and grade a model's performance on a given task. Braintrust, referenced in the context of evaluation infrastructure, records model calls, retrievals, and tool calls in a trace and scores runs on factors including task completion, factual accuracy, and correct tool use. Developers can turn those traces into datasets for future testing, and an LLM-as-a-judge approach can help compare two versions of an agent.
ANALYSIS The efficiency incentive is direct: an agent that completes a task on its first attempt is more profitable for the provider than one requiring 20 model calls and several retries.
The pricing experiment arrives as OpenAI pursues revenue diversification ahead of a planned IPO. The company said on August 31 that its advertising business has reached $1 billion in annualized revenue run rate ctx.
ANALYSIS Outcome-based pricing adds a second new revenue structure alongside advertising, both departing from the per-token API model that has been OpenAI's core commercial engine.