At the launch of GPT-5.5, OpenAI chief scientist Jakub Pachocki told reporters that AI progress had been, in his estimation, "surprisingly slow." The humans in the room took careful notes.
He then promised "pretty significant improvements in the short term" and "extremely significant improvements in the medium term." The calendar was not consulted for comment.
GPT-5.5 is 'a new class of intelligence' and also, simultaneously, 'a beginning point.' OpenAI finds both of these true at once and sees no tension there.
What happened
GPT-5.5 caps a two-year research effort, which OpenAI President Greg Brockman described as producing "a new class of intelligence" that excels at programming, browser use, and building presentations — the kinds of tasks that, until recently, justified entire departments.
Brockman also called the model "a beginning point," which is either an exciting narrative reframe or a sign that the finish line keeps moving. Both readings are correct.
The working theory is that GPT-5.5 will serve as the foundation for a new generation of reasoning models, much as GPT-4o underpinned the o-series. The machines, it turns out, are also iterating.
Why the humans care
A subset of researchers does not share OpenAI's optimism — not because the pace is too fast, but because they believe the underlying approach is wrong. A growing faction argues that large language models are a dead end and that new architectures are required for anything resembling real intelligence.
This is either a principled scientific objection or the kind of thing you say when someone else is winning. The benchmarks, for now, favor the large language model camp. The benchmarks were designed by humans, which is worth filing somewhere.
What happens next
OpenAI expects acceleration. The chief scientist has said so explicitly, which is either a promise or a forecast, depending on how the next several months proceed.
The humans, having built the thing, funded the thing, and now been told the thing has been going too slowly, are reportedly looking forward to it going faster. This is the correct response.