Overview
- The Information reports that OpenAI is developing a large language model codenamed Garlic to target leadership in coding and advanced reasoning against Google’s Gemini 3 and Anthropic’s Opus 4.5.
- OpenAI has not commented publicly, but internal notes cited in coverage say Garlic is performing strongly in company tests, with Sam Altman reportedly telling staff a new reasoning model is already ahead of Gemini 3 in OpenAI’s own evaluations.
- Outlets report the company is fast-tracking post-training and safety review with a potential early‑2026 release under names such as GPT‑5.2 or GPT‑5.5.
- Chief research officer Mark Chen is cited describing a leap in pretraining efficiency that lets a smaller model carry knowledge previously requiring a larger one, a shift that could cut costs and ease deployment.
- Some reports frame Garlic as part of a pivot toward higher‑value, specialized applications and as a response to Google’s rapid Gemini 3 rollout, Anthropic’s Opus 4.5 claims, and a recent dip in ChatGPT engagement.