Interesting： OpenAI says GPT-5.6 Sol helped cut its end-to-end model-serving costs by 20%， by autonomously rewriting and optimizing production GPU kernels.

Sol also improved its own speculative decoding model： - Designed and ran hundreds of architecture experiments - Launched and monitored the training process - Intervened during hardware failures and training instability

The resulting system increased token-generation efficiency by more than 15%.