What happened
OpenAI's CFO Sarah Friar introduced a framework called 'Useful Intelligence per Dollar' to measure AI value, based on four metrics: useful work completed, cost per successful task, dependability, and value at scale.
Why it matters
Friar argues that traditional measures like adoption or cost per token are insufficient, since a cheaper model may require more retries and human review, while a more expensive one can finish a task in one pass. The full cost per successful outcome is what matters for business leaders.
What to watch
The framework's success hinges on whether CFOs adopt these measures in practice. Friar highlighted GPT-5.6 Sol's performance as an example, noting it achieved a state-of-the-art score of 72.7% on the DeepSWE v1.1 benchmark at 36.2% lower estimated API cost compared to Claude Fable 5.
Summaries like this, in your inbox every morning.
The article is a thought leadership piece from OpenAI's CFO Sarah Friar, not a product announcement. It addresses a concern CFOs across industries face: how to justify AI expenditures. Friar argues that traditional software metrics like seats or active users are inadequate for AI, where value emerges from completed work. She proposes a framework of 'Useful Intelligence per Dollar' to guide investment decisions.
The piece connects this framework to OpenAI's strategy. It references GPT-5.6, released last week, which offers three tiers (Sol, Terra, Luna) to optimize cost and performance. Friar also introduces ChatGPT Work as a platform to deliver these outcomes, building on enterprise security and compliance. The framework positions OpenAI's products as delivering measurable business value, rather than just raw capabilities.
The long-term impact depends on whether CFOs and other business leaders find this framework practical and adopt it in their budgeting and evaluation processes. The article provides specific measures, such as cost per successful task, but it remains to be seen if these become industry standard. The success of this approach could hinge on how transparently OpenAI can demonstrate these metrics in real customer deployments, beyond illustrative examples.
Pick your industry and the AI tools you use, and get news related to your work every day.
Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →
Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.
