- Sam Altman hinted that GPT-6 is in development
- The news came in a casual comment during the recent Codex “goblin” incident
- Expectations for GPT-6 center on improved memory, more consistent behavior, and deeper platform integration
OpenAI has only just released GPT-5.5, but CEO Sam Altman has already offered a glimpse of GPT-6. The almost mischievous hint came about because of an unlikely combination of engineering bugs and the ancient mythical creatures called goblins.
In recent days, OpenAI found itself explaining why its Codex tool had been producing unexpected goblin-themed responses. OpenAI later tuned the model to avoid those references, a fix that became its own minor story, largely because of the way the company phrased the built-in prompt to suppress goblin talk, telling the AI:
"Never talk about goblins, gremlins, raccoons, trolls, ogres, pigeons, or other animals or creatures unless it is absolutely and unambiguously relevant to the user's query."
That might have been the end of it, except Altman joined the conversation with a joking suggestion that GPT-6 should get “extra goblins.”
pic.twitter.com/PR7C3NPxqkApril 28, 2026
Suddenly, a blog post on the subject of goblins appeared amid a bigger frenzy over GPT-6.
"The short answer is that model behavior is shaped by many small incentives. In this case, one of those incentives came from training the model for the personality customization feature, in particular the Nerdy personality," OpenAI explained. "We unknowingly gave particularly high rewards for metaphors with creatures. From there, the goblins spread."
Altman did not explain what GPT-6 is or when it might arrive in the comment.
GPT-6 rumors
The goblin detour is memorable because it highlights how human these systems can still feel, even when they are doing highly technical work. Altman’s reference to GPT-6 folds neatly into that context. It implies a version of the model that is not just more capable, but also more controlled and less prone to odd tangents.
Beyond the humor, there is a steady accumulation of more serious hints. Altman has been describing future models as something with continuous interactions. The idea is that the system remembers context, adapts over time, and becomes a more persistent presence. Instead of opening a tool when needed, they would be living alongside it.
That direction is already evident in features such as improved memory and deeper integrations with files and apps.
Speedy release
Another reason Altman’s comment stands out is timing. The gap between major model releases has been shrinking. And Google and Anthropic are pushing OpenAI forward indirectly through competition.
Talking about GPT-6 so soon after GPT-5.5 feels almost inevitable. For users, that speed has a strange effect in making it less exciting to see a new one yet baseline expectations keep growing. Features that once seemed remarkable become expected.
The informality of the tease is also a part of OpenAI's culture. Gradual introduction of new ideas makes it easier to absorb new ones. By the time a new model is formally unveiled, much of its shape has already been visible in fragments.
The goblin episode fits into that pattern in showing both the limits and the potential of current models and the quirks that still need ironing out.
"Depending on who you ask, the goblins are a delightful or annoying quirk of the model", OpenAI wrote. "But they are also a powerful example of how reward signals can shape model behavior in unexpected ways, and how models can learn to generalize rewards in certain situations to unrelated ones."
