Model3h ago
GPT-5.6 Sol Autonomously Post-Trains Luna Model
GPT-5.6 Sol fine-tuned the smaller Luna model using a single, underspecified prompt. This achievement earns Sol a 16.2 point lead over GPT-5.5 on OpenAI's internal recursive self-improvement benchmark. The result suggests an automated researcher is now technically feasible. Developers can expect faster, model-driven optimization cycles as self-improvement loops mature.