OpenAI outlines a framework called "Useful Intelligence per Dollar" to help businesses measure the economic value of their AI investments beyond simple cost-per-token metrics. The article argues that success should be measured by work accomplished, reliability, and scalability rather than just adoption rates.
The scorecard evaluates four key areas: whether AI completes meaningful work, the full cost per successful task (including retries and human review), dependability through quality tracking, and whether value improves at scale. OpenAI highlights its recently released GPT-5.6 model family, which includes Sol, Terra, and Luna tiers, as an example of optimizing this equation. On the Artificial Analysis Coding Agent Index, GPT-5.6 Sol with max reasoning set a new state of the art while using 54% fewer output tokens than a leading competitor. The framework emphasizes that lower-cost models may require more attempts, whereas capable models can deliver results in one pass, reducing total compute and review time.
OpenAI positions this metric as a way for CFOs to determine if AI value grows faster than production costs, aiming to make AI more useful by lowering the cost of work while increasing capability.