Altman: Codex finished coding at naptime, teases ‘Goblin’ name

Sam Altman wrote on X that OpenAI may call its next model “Goblin” and that he returned from time with his child to find a series of Codex tasks finished. He posted the account as an example of the coding system running without human supervision and added a separate post floating the name suggestion.

Altman posted a message describing how he had started “a bunch of codex tasks,” left to spend time with his child and came back at naptime to find them all completed. He added that the experience made him “very optimistic for the future.” He later wrote, “what if we name the next model ‘goblin’ — almost worth it to make you all happy.”

Codex is OpenAI’s tool that converts natural-language prompts into working code. The company now promotes it as an agent that can hold a task list, sequence steps and return finished outputs without a developer watching every action. Altman’s anecdote was presented alongside that shift in how Codex is marketed.

OpenAI has pushed autonomous workflows in its enterprise pitch and has expanded cloud reach through a partnership with Microsoft. Competing firms are developing similar hands-off coding assistants intended to complete multi-step tasks with limited human oversight.

Altman also ran a poll on X asking users what they wanted improved in the next model. He reported that the poll results closely matched OpenAI’s roadmap and noted that many respondents asked for “more goblins.” OpenAI published a report titled “Where the Goblins Came From” on April 29 that examined why recent models began using goblins, gremlins and similar creatures in metaphors. The report said training for personality customization, particularly a “nerdy” or metaphor-rich voice, unintentionally rewarded creature-based metaphors and contributed to the spread of “goblin” references.

Earlier posts from Altman characterized the current model as an “autistic genius,” a phrase used to describe strong technical performance paired with uneven tone in other contexts. OpenAI has said it is adjusting training incentives for personality customization while continuing to develop tools designed to complete complex workflows with less human supervision.

The company has not announced a release date for the next model. The naming idea and the Codex anecdote have drawn attention to both the tool’s autonomous capabilities and the models’ emerging language patterns.