OpenAI Tests Outcome-Based Pricing for Enterprise Customers

OpenAI has begun testing a new billing arrangement with select enterprise customers that charges only when its AI systems successfully complete tasks, according to a report by The Information's Kevin McLaughlin and Amir Efrati. The pilot program represents a departure from the company's standard token-based pricing model.

The arrangement remains limited to major accounts and has not been publicly announced. Details about participating customers, specific pricing, and contract terms have not been disclosed.

Why it matters

Outcome-based pricing addresses a fundamental problem in enterprise AI adoption: unpredictable costs that scale with attempts rather than results. When bills arrive regardless of whether AI systems produce useful work, finance leaders struggle to justify budgets. This shift transfers risk from buyers to vendors and could accelerate enterprise AI deployment by removing a major barrier at the contract negotiation stage.

The token billing problem

Under traditional token-based pricing, customers pay for API calls and compute usage whether or not the AI produces valuable results. One developer reportedly accumulated $1.3 million in OpenAI charges over thirty days while running a hundred agents in parallel, illustrating how costs can spiral with experimentation and failed attempts.

This consumption model creates forecasting challenges that often stall enterprise deals. Companies running pilots find it difficult to predict production costs, leading to extended evaluation periods rather than committed contracts.

Customer support leads the shift

Outcome-based pricing has already taken hold in AI-powered customer support, where success can be clearly defined as a resolved conversation. Intercom charges $0.99 per resolution its Fin agent completes and nothing for conversations it cannot handle. Zendesk refined this further in May with "Verified Resolutions" confirmed by LLM evaluation within 72 hours, charging approximately $1.20 to $1.50 per confirmed resolution on volume commitments.

Salesforce has navigated this transition more publicly. Its Agentforce product initially charged $2 per 24-hour conversation session regardless of outcome, drawing criticism for high costs and poor predictability. The company responded with Flex Credits, moving to action-based pricing at roughly 10 cents per action with a $500 minimum for 100,000 credits. However, this remains consumption pricing rather than outcome pricing, since failed actions still generate charges.

Market demand and vendor risk

Research from Futurum Group in May found that 43 percent of enterprise buyers prefer consumption-based models and 27 percent prefer outcome-based pricing, with fewer than one in five still favoring per-user licensing. Research director Keith Kirkpatrick noted that vendors offering only seat-based pricing are now being eliminated from consideration before evaluations begin.

For vendors, outcome-based pricing requires confidence in success rates. The risk of failed attempts must be absorbed into per-success pricing, which explains why resolution rates cluster around one dollar rather than a few cents. The model also requires clear definitions of success, which works for customer support resolutions but becomes more complex for multi-step agentic tasks where completion involves judgment calls.

OpenAI's move into outcome-based pricing, while limited and unannounced, signals that even the largest model providers recognize the need to align billing with business value rather than infrastructure consumption.

The Information first reported these details about OpenAI's outcome-based pricing pilot.