OpenAI proposes 'Useful Intelligence per Dollar' framework, citing a cost-adjusted coding lead over Claude Fable 5

OpenAI published a strategic framework this week for judging AI progress by economic value rather than raw capability, arguing that the right yardstick for enterprises is what it calls "Useful Intelligence per Dollar" — and using GPT-5.6 Sol's coding-benchmark results against Claude Fable 5 to illustrate the point.

What's new

In the post, OpenAI frames the core business question plainly: "The basic economic question facing CFOs and other business leaders is whether the value of the work AI completes grows faster than the cost of producing it." The company backs the framework with a specific comparison: it says GPT-5.6 Sol scored 72.7% on the DeepSWE v1.1 benchmark, ahead of Claude Fable 5's 69.9%, while running at an estimated 36.2% lower API cost. OpenAI also cites a separate efficiency figure — GPT-5.6 using 54% fewer output tokens than a leading rival model on the Artificial Analysis Coding Agent Index — as further evidence for its cost-per-outcome argument.

Context

The post comes roughly a week after GPT-5.6's July 9 launch and amid intensifying competition on cost-efficiency claims across frontier labs, as rivals including Anthropic and Alibaba's Qwen team have made similar arguments about price-to-performance in coding and agentic workloads. OpenAI's framing shifts the debate away from leaderboard position alone and toward what it calls the "full cost of producing a successful outcome, measured against the value that outcome creates."

Why it matters

Benchmark headlines increasingly obscure cost, and OpenAI's framework is an explicit attempt to reset the comparison on its own terms — using a metric it defines and a benchmark pairing it selected against a direct competitor's flagship model. Enterprises evaluating which model to standardize on for coding and agentic work will need cost-adjusted numbers like these regardless of which lab publishes them, and this signals OpenAI expects future competitive claims to be litigated on economics, not just accuracy. As OpenAI puts it: "Capability earns first use. Dependability makes AI part of how work gets done."