OpenAI allocates 80 to 90 percent of its research budget toward future frontier generations like GPT-7 and GPT-8, according to Boris Power, the company’s Head of Applied Research. Speaking on the company’s internal strategy, Power noted that major generational shifts provide the vast majority of technical and commercial value compared to incremental updates.
While intermediate updates like GPT-5.1 or 5.2 help the company iterate quickly using specialized training data, Power characterized these short-term improvements as inherently limited. After every major release, the company must reassess where to invest for immediate wins, but long-term breakthroughs require fundamental model advances where overall capability increases across all domains.
Power also highlighted onboarding and user interaction as the primary bottleneck for current AI assistants. As model capabilities progress from GPT-4’s prompting requirements to GPT-6’s autonomous goal execution, future systems must better anticipate user needs and proactively surface capabilities without requiring detailed instruction.
Why it matters
AI labs are focusing resources on next-generation architectures rather than optimizing current model iterations.
Product teams should design for proactive agentic workflows as future models shift from prompt response to goal delegation.
Incremental model updates offer diminishing returns compared to generational leaps in baseline intelligence.
Source: the-decoder.com



