OpenAI proposes new AI ROI metric: value per dollar

OpenAI's announcement on July 17, 2026 aims to shift how businesses evaluate AI investments. The new framework moves beyond the industry's common focus on token pricing, pushing companies to consider the full economic picture of a task. By defining "output" in business terms—like customer support ticket resolutions or code changes merged—the company wants organizations to measure the actual productive output AI enables, not just the cost per token.

The framework also highlights a practical insight: a cheaper model that requires multiple retries and human review can end up costing more than a more expensive model that gets the answer right the first time. This suggests that the total cost of a completed task—including compute, labor, and rework—should guide model selection. OpenAI's proposed hybrid approach, using lightweight models for routine work and top-tier models for complex reasoning, could help companies balance cost and performance.

The emphasis on reliability and trust is notable. OpenAI outlines a three-level pattern for evaluating trust: output usable as-is, output needing editing, and output requiring escalation. This acknowledges that AI adoption is gradual, and human oversight remains necessary. As companies scale AI use, tracking metrics like task volume, cost, and value over time could become standard practice, potentially changing how AI ROI is assessed across industries.