Anecdotally, as someone who operates in a very large legacy codebase, I am very impressed by GPT-5's agentic abilities so far. I've given it the same tasks I've given Claude and previous iterations via the Codex CLI, and instead of getting loss due to the massive scope of the problem, it correctly identifies the large scope and breaks it down into it's correct parts and creates the correct plan and begins executing.
I am wildly impressed. I do not believe that the 0.x% increase in benchmarks tell the story of this release at all.
I'm a solo founder. I fed it a fairly large "context doc" for the core technology of my company, current state of things, and the business strategy, mostly generated with the help of Claude 4, and asked it what it thought. It came back with a massive list of detailed ambiguities and inconsistencies -- very direct and detailed. The only praise was the first sentence of the feedback: "The core idea is sound and well-differentiated."
I am wildly impressed. I do not believe that the 0.x% increase in benchmarks tell the story of this release at all.