OpenAI Uses GPT-5.6 Sol to Train Its Own Sibling Model, Eyes Fully Automated AI Researcher by 2028

OpenAI's GPT-5.6 Sol is now training its sibling model GPT-5.6 Luna by performing senior-level research tasks, marking a major step toward the company's ambitious goal of deploying a fully automated AI researcher capable of end-to-end research with minimal human oversight by March 2028.

• OpenAI's flagship GPT-5.6 Sol model is being used to handle post-training for its lightweight GPT-5.6 Luna model, performing senior-level research tasks including setting up training runs, monitoring jobs, repairing technical problems, running evals, and overseeing workflows.
• Sol's involvement marks a significant leap beyond previous model-assisted training efforts, surpassing what an entry-level researcher could handle, though it stops short of recursive self-improvement and is not replacing human researchers but rather freeing them to focus on more creative work.
• OpenAI is now looking to apply these learnings across its entire research pipeline, with ambitions extending toward a fully automated AI researcher by March 2028 capable of handling complete research tasks end-to-end with minimal human interaction.