GPT-5.6 Sol: AI autonomously optimizes Luna model

OpenAI's new AI model, GPT-5.6 Sol, has autonomously post-trained a smaller model called Luna. Sol independently optimized Luna with just a "fairly underspecified prompt." What's interesting is that Sol identified training configurations, selected GPUs, and executed the post-training script all on its own. This capability sets a new benchmark for AI self-improvement. On an internal benchmark measuring recursive self-improvement, GPT-5.6 Sol scored 16.2 points higher than its predecessor, GPT-5.5. An OpenAI employee noted that while Sol adapted an existing setup, this autonomous task would have otherwise taken two staff researchers about two weeks. The bottom line is that this development shows AI models are becoming more capable of improving themselves, potentially speeding up AI development significantly.