OpenAI's GPT-5.6 Sol can train smaller models on its own, boosting performance by 16.2%

OpenAI's new GPT-5.6 Sol model can automatically train smaller models like Luna. This process, called post-training, adjusts a model after its initial training. In tests, GPT-5.6 Sol scored 16.2% higher than GPT-5.5 on tasks like debugging and running experiments. Researchers say it acts like an automated helper, reducing the need for human work. OpenAI reports that developers using GPT-5.6 Sol have doubled their daily token output and increased experiments and code submissions.