GPT-5.6 Sol helped optimize the infrastructure that serves it

OpenAI reports using GPT-5.6 Sol to improve the software that runs its models. Its engineering account describes the model finding avoidable computation and inefficient data movement, then rewriting GPU kernels, the routines that execute model mathematics.

The company attributes a 20% reduction in end-to-end serving costs to that work combined with broader kernel improvements. It also describes verification tooling used to check generated kernels for correctness. This is an account of model-assisted infrastructure engineering: the reported gain concerns the cost of operating the service, rather than a measured improvement in answer quality or a universal reduction in customer prices.

“Our primary objective is to serve more tokens with the same hardware, while preserving the intelligence, latency, availability, and reliability users expect.”

“If a task requires 30 model requests, an extra second per request adds up.”

“Tool output is capped at 10,000 tokens by default unless the model requests a different limit.”