OpenAI directs 80% to 90% of its research toward GPT-7, GPT-8 and later systems, according to Boris Power, the company's head of applied research. His comments show how far ahead frontier-model development runs while customers are still learning to use the current generation.
Power described updates within one generation, such as moving from GPT-5.1 to GPT-5.2, as short-term efforts built around specialized training data. They can improve today's product and teach the company where quick gains are available, but he said the larger value comes when a new model generation improves many capabilities at once.
He also argued that raw model quality is no longer the biggest obstacle for many ChatGPT users. People often do not know which tasks an assistant can handle or how to delegate work effectively. In his account, GPT-4 required careful prompting, GPT-5 reduced that burden and GPT-6 behaves more like a colleague that can receive a goal.
These are statements from an OpenAI executive, not an independent comparison of the models. They indicate the company's research allocation and product priorities, but provide no release schedule or performance evidence for GPT-7 and later systems.