GPT-5.2: Post-training RL and long-horizon reasoning

OpenAI’s GPT-5.2 distinguishes itself through post-training reinforcement learning, prioritizing longer-horizon, agent-like reasoning as pre-training performance among models converges.