OpenAI Directs 90 Percent of Research to GPT-7 and Beyond

OpenAI is heavily prioritizing long-term frontier model development over minor, iterative upgrades. Speaking about the company's internal roadmap, Boris Power, OpenAI's Head of Applied Research, revealed that between 80 and 90 percent of the organization's research resources are currently dedicated to GPT-7, GPT-8, and even more advanced future systems. According to Power, the company focuses on these distant horizons because that is where the vast majority of the technology's ultimate value lies.

While the company continues to release mid-generation updates, such as moving from GPT-5.1 to GPT-5.2, Power characterized these incremental improvements as "extremely shortsighted" when viewed as a long-term strategy. These minor steps typically rely on specialized training data and serve as temporary measures to help the company iterate and learn in the present. However, the true breakthroughs occur during major generational shifts, which fundamentally upgrade how the entire system functions and force the research team to relearn where to find quick wins.

For developers and practitioners, this long-term focus will radically change how humans interact with AI. Power mapped out this evolution, noting that GPT-4 required highly specific and careful prompting to produce useful results. The upcoming GPT-5 is designed to be easier to use, though it still requires significant user feedback to guide its outputs. By the time OpenAI reaches GPT-6, the system is expected to function like a capable colleague to whom a practitioner can simply hand a broad goal and expect autonomous execution.

This shift aims to solve what Power identifies as the primary bottleneck for current AI assistants: user onboarding rather than raw model quality. Because many current ChatGPT users struggle to understand the full capabilities of the technology, future model generations must become better at anticipating user needs and proactively demonstrating what they can achieve.