OpenAI’s About $40 Billion Revenue Run Rate Highlights AI Pricing Divide

OpenAI’s about $40 billion annualized revenue pace in July 2026 underscores how far AI sales have scaled, while Codex billing keeps charges tied to model use rather than customer outcomes.

OpenAI Chief Financial Officer Sarah Friar put the company’s annualized revenue pace at about $40 billion in July. She said enterprise subscriptions and API sales accounted for more than half. Business spending on OpenAI’s API then grew 82% from the second quarter to the third quarter.

The pricing question sits alongside measurable progress in AI tasks. GPT-5.5 matched or surpassed industry experts on 84.9% of 44 GDPval professional tasks. The model scored 78.7% on OSWorld-Verified, above the test’s human baseline. Friar also described Travelers’ AI claims assistant expanding from eight states to nationwide coverage in two months.

Those results show gains on specific work, but a benchmark score or a claims tool’s expansion does not measure whether AI can manage an entire company. Running a business also calls for choices amid incomplete information, oversight of cash flow, responses to regulation and customers, and decisions whose consequences can be difficult to reverse.

Codex’s billing change in April 2026 illustrates how OpenAI charges for access to its tools. The service moved from estimated message-based charges to billing aligned with API token consumption. That ties the fee to the amount of model use, regardless of whether the customer’s wider business objective is achieved.

Friar has described outcome-based pricing as a future direction for OpenAI and argued that businesses should assess AI by the cost of a task rather than the cost of each token. Those measures can tell a buyer what it costs to complete a defined job. They do not, on their own, settle who is responsible when an AI system misses a broader business goal.

The distinction is between paying for activity and paying for a result that can be counted. Under token-based billing, the customer’s charge follows the model’s use. A fee tied to a successful result shifts some exposure to the provider, while requiring the parties to agree on what success means and how to verify it.

HubSpot introduced outcome-based fees for two AI services in April 2026. Its Breeze Customer Agent costs 50 cents for each successfully resolved customer-service conversation, down from $1 per conversation. Its Breeze Prospecting Agent costs $1 for each lead it recommends for outbound calls.

Those fees attach to discrete events: a customer conversation is resolved or a lead is recommended. The measure is narrower than the performance of a company, which can depend on market conditions, spending decisions, customer relationships and regulatory obligations. The examples therefore show a way to price specific AI tasks, rather than a model for transferring responsibility for an entire business.

OpenAI’s revenue figures and API spending growth show the scale of selling access to AI tools. The benchmark results and Travelers example show systems handling defined work in business settings. Together, they frame the gap between completing a task and taking responsibility for a company’s overall performance. Outcome-based fees have appeared in bounded services, while Codex billing remains linked to token consumption.