OpenAI’s Next Model Could Be Named “Goblin,” Sam Altman Hints

Sam Altman said launching a batch of Codex coding jobs left him optimistic about autonomous agents. He returned at his child’s naptime to find every task already finished.

Just one day after describing OpenAI’s model as an “autistic genius,” Sam Altman joked that the next AI release should be named “Goblin.”

Codex Handles Tasks Without Supervision

Codex is OpenAI’s coding system. It interprets natural-language prompts and produces working code without a developer reviewing each step. Altman’s anecdote signals that the tool can now run unattended on real work.

“kicking off a bunch of codex tasks, running around with my kid in the sunshine, and then coming back at naptime to find them all completed makes me very optimistic for the future,” – Altman, X

OpenAI’s positioning has moved with it. Codex is no longer pitched as an autocomplete tool. The company markets it as an agent that holds a task list, sequences its steps, and returns finished output.

That puts Codex against rival coding assistants from Anthropic and Google. Both companies are racing to deliver the same hands-off workflow. OpenAI’s enterprise pitch leans on these autonomous flows, especially after the company broadened its cloud reach with Microsoft.

Goblins, Genius, and the Next Model

In the same window, Altman polled X users about what they wanted improved in OpenAI’s next model. He later said the results “reasonably well” matched the roadmap, then flagged the popular request for “more goblins.”

Hours later, he wrote that naming the next release “Goblin” was “almost worth it to make you all happy.” OpenAI had already published a report on April 29 titled “Where the Goblins Came From.” It examines why models starting with GPT-5.1 began reaching for goblins, gremlins, and similar creatures in their metaphors.

The short answer is that model behavior is shaped by many small incentives, and in this case, one of those incentives came from training for personality customization, especially a “nerdy” style where metaphorical and creature-based language was unintentionally rewarded, which is how the “goblins” spread.

The “autistic genius” line, used earlier on X, frames the current model as powerful but uneven. The contrast shows how OpenAI’s flagship system can outperform humans on technical tasks yet stumble on tone.

The two threads sit unclearly together. OpenAI is showing Codex completing serious engineering work without supervision. Its consumer model, meanwhile, has developed a folkloric vocabulary that researchers needed to investigate. The next release will reveal whether autonomous tooling or model personality drives the headline for OpenAI in 2026.