Inside OpenAI’s Hidden ‘Garlic’ Model That’s Challenging Google Again
OpenAI’s secret “Garlic” AI aims to match Google’s Gemini 3, with internal tests showing strong coding and reasoning scores ahead of an early 2026 launch.
image for illustrative purpose
OpenAI is working on a novel AI system which is reportedly called “Garlic” internally with a rollout planned for early 2026, as the rivalry with Google heats up.
A piece by The Information disclosed that OpenAI’s Chief Research Officer, Mark Chen, informed employees that not only is the model in development but it is also carrying out reasoning and coding tasks like Google’s Gemini 3 and Anthropic’s Opus 4.5 in the measuring tests of the internal performance. The company is considering using either the GPT-5.2 or GPT-5.5 tag for the new system.
With the company’s CEO Sam Altman directing the teams this week to boost ChatGPT’s core intelligence and to put on hold the development of other consumer products including health and shopping tools the urgency of the project just escalated. Altman also mentioned that OpenAI has a new reasoning engine which is better than Gemini 3 in the internal benchmarks.
The report also mentioned that even if the Garlic model had early success, it would still have to undergo post-training refinements, evaluations, and standard safety checks before it can be made available to the public.
Throughout this time, Google has been gaining attention for its launch of Nano Banana Pro and its advanced Gemini 3 series. The company’s Gemini application has recently experienced an increase in user engagement, receiving compliments not only from users but also from the likes of Salesforce CEO Marc Benioff.
