OpenAI's GPT-5 delivers incremental gains, not AGI leap

Nathan Lambert's analysis of GPT-5 reveals a model that is both impressive and underwhelming, depending on expectations. OpenAI's flagship was hyped as a potential AGI milestone, but business realities forced it to be a practical, scalable product rather than a superintelligence. The release was messy—mislabeled plots, live demo bugs, and a confusing rollout—reflecting OpenAI's internal tension between satisfying AI insiders and serving the mass market. GPT-5 tops LMArena in all categories and leads the ArtificialAnalysis composite benchmark, but these gains are incremental rather than revolutionary. It matches best-in-class models like Claude Sonnet for coding and o3 for search, funneled into one cheap package.

The core implication for the next 6–18 months is that AI progress is shifting from raw performance to a multifaceted race involving price, product integration, and user experience. GPT-5's modest step on the performance trend line suggests that the field is entering a more traditional technological path, where many small improvements accumulate over time. Lambert notes that the most important day in AI may be the release of an open model alongside GPT-5, not GPT-5 itself. For professionals, this means that while GPT-5 won't revolutionize reasoning overnight, it will make AI more accessible and reliable for everyday tasks, solidifying OpenAI's market position as the industry moves toward commoditization.