AI Pricing Models 2026: OpenAI Outcome-Based Pricing vs Per-Token

OpenAI has quietly started letting some of its largest enterprise customers pay only when its AI actually completes the job — the reported example is a customer-support interaction handled end-to-end — instead of paying per token or per API call. The arrangement is limited to select major accounts, not generally available, and OpenAI has not formally announced it; the terms, the customers, and the prices are all unknown. The Information first reported the shift on August 30–31, 2026; TNW, PYMNTS, and Yahoo Finance carried it the same day.

For AI agencies this is the pricing story of the year: the vendor that built the token economy is now selling results, not capacity. If you quote retainers, pass through token costs, or price per task, this changes how you model margin and what you can promise clients.

What OpenAI is doing

Four facts are worth pricing into every agency model:

1. OpenAI has begun letting some of its largest enterprise customers pay only when its AI completes the job — for example, a customer-support interaction handled end-to-end.
2. For those accounts, it replaces per-token and per-call pricing — a shift away from conventional usage-based pricing.
3. The arrangement is limited to select major accounts. It is not generally available.
4. OpenAI has not formally announced it. TNW notes it has not independently verified the report, and "the terms, the customers, and the prices are all unknown."

TNW's framing is the one to quote in client conversations: "What is on the record is that the largest model vendor has started, quietly and selectively, to sell results instead of capacity." This is separate from the revenue-share pricing OpenAI's CFO floated in January.

Outcome-based pricing was already the standard in support AI

OpenAI is late to a party it helped make possible. Outcome-based pricing is already how customer-support AI is sold, and the reference prices are public:

Buyers are already voting: Futurum Group's 1H 2026 survey found 43% of enterprise buyers prefer consumption-based models, 27% outcome-based, and fewer than 20% per-user. As research director Keith Kirkpatrick put it: "Outcome-based pricing is becoming a market standard … vendors offering seats alone are now being disqualified before the evaluation starts." Salesforce CEO Marc Benioff said last week: "Customers want to buy and want to price in different ways. This is something I've learned really aggressively recently."

Two things make OpenAI's move bigger: it is the model vendor whose per-token pricing created the usage-based economy agencies bill against, and the platform itself — the same APIs you build on can now be metered by completed work.

What changes for agencies: retainer vs usage vs outcome

AI pricing models 2026 run on three structures: retainer (fixed fee), usage-based (per token, per call), and outcome-based (per completed job). Our agency pricing models explainer walks through all three; this news changes the balance between them.

The risk transfer is the headline. Under token billing, the customer pays for every failed attempt, retry, and loop. Under outcome billing, the vendor absorbs failures — a material transfer of risk. A vendor less confident of success has to load the per-success price until the economics match, which is why per-resolution rates cluster around a dollar; agencies quoting outcomes must bake in a retry rate or their margin disappears. The extreme case TNW cites: one developer running 100 agents in parallel reported a $1.3M OpenAI bill in 30 days — cost scales with attempts, not results.

• The "unforecastable bill" objection disappears where pilots stall. If you resell API capacity, outcome pricing replaces the client's fear of an open-ended token bill with a per-completed-task price you can quote with confidence.
• Model per-completed-task vs API usage side-by-side. That is exactly what the outcome-pricing mode on this calculator runs: completed tasks per month, price per outcome (market reference $0.50–$2.00 per resolution), and an optional per-token equivalent for the same work.
• Define "success" in the contract. Support resolutions are definable; agentic work involves multi-step tasks where completion is a judgment call. Stripe's outcome-pricing guidelines warn: "Unless attribution rules are explicit, customers could argue over whether the outcome belongs to you."
• Watch the margin math at high volume. Outcome pricing is cheaper when success rates are low and riskier when the per-outcome price exceeds your effective token cost.

Should you sell outcomes, not services?

The strategic play is not to wait for OpenAI to offer outcome pricing to your clients — it is to offer it yourself. Our Sell Outcomes, Not Services page makes the case: price the outcome, track the cost underneath it, and keep the margin when unit prices fall.

The Jevons-paradox data already on this site shows why. When OpenAI discounted two GPT-5.6 models in July–August 2026, usage on the most-discounted model grew 13.8x — cheaper tokens did not mean smaller bills, they meant far more usage. Price per token and falling model prices shrink your bill; price per outcome and they expand your margin. That is the structural argument for outcome-based agency pricing.

Frequently asked questions

Sources

• TNW, "OpenAI has started letting some customers pay only when the AI works" (Aug 31, 2026): thenextweb.com/news/openai-outcome-based-pricing-enterprise
• The Information, "OpenAI Starts Letting Some Customers Pay Only When the AI Works" (Kevin McLaughlin / Amir Efrati, Aug 30–31, 2026): theinformation.com
• PYMNTS, "OpenAI Lets Some Customers Pay Only When AI Performs" (Aug 31, 2026): pymnts.com
• Yahoo Finance / StockTwits, "OpenAI Reportedly Joins Salesforce, Others In Testing Outcome-Based AI Pricing" (Aug 31, 2026): finance.yahoo.com
• Futurum Group, "Are outcome-based and hybrid AI pricing models rewriting the vendor playbook?" (May 12, 2026): futurumgroup.com
• Korix, "The 6 AI Pricing Models, Ranked by Real 3-Year Cost" (2026; secondary vendor-published data for HubSpot/Salesforce figures): korixinc.com
• Sierra, "Outcome-based pricing for AI agents" (May 12, 2026): sierra.ai

Accuracy note: All facts, quotes, and figures verified against the research brief (t_9974fbbc, evidence gate GREEN, 17 sources / 25 verbatim quotes) on 2026-08-31. The Information is paywalled; its reporting is attributed via TNW, PYMNTS, and Yahoo Finance coverage of the same report. No customer names, prices, or launch date are public. HubSpot ($0.50) and Salesforce list-price figures are vendor-published/secondary data and cited as such. OpenAI's outcome-based pricing is limited to select enterprise accounts and is not generally available — re-verify before quoting client work.