GPT-5.2 and the Shift to Post-Training RL

As model pre-training converges, OpenAI's GPT-5.2 uses post-training RL to differentiate through long-horizon, agentic reasoning rather than standard demo performance.