OpenAI says GPT-5.6 Sol autonomously post-trained Luna

OpenAI disclosed that GPT-5.6 Sol autonomously post-trained the smaller Luna model.

The task included finding training configurations, selecting GPUs, launching the training script and verifying the run.

OpenAI built an internal evaluation suite around real AI research tasks such as debugging systems and running experiments.

GPT-5.6 Sol scored 16.2 points higher than GPT-5.5 on the aggregated RSI index, though staff noted it adapted an existing configuration rather than inventing one from scratch.