OpenAI Is Developing A New Model Called Garlic

Image on Unsplash

OpenAI is developing a new large language model called Garlic that is designed specifically to rival Google's Gemini 3 and Anthropic's Opus 4.5. It will be a competitor model in advanced reasoning and coding capabilities, in particular.

The announcement shows that OpenAI is not just responding to Google's competitive threat by improving existing models. They are building an entirely new model architecture from scratch specifically designed to beat what Google has already accomplished.

Early internal testing of Garlic suggests the model is performing strongly across multiple benchmarks and could potentially launch as GPT-5.2 or GPT-5.5 by early 2026. This would represent a faster iteration cycle than OpenAI has traditionally maintained.

Just days ago Sam Altman declared code red at OpenAI over Google's advances with Gemini 3. Now it emerges that OpenAI had already been working on Garlic well before that public announcement, suggesting the internal pressure to compete with Google has been building for months.

The focus on reasoning and coding capabilities is crucial because these are the areas where Gemini 3 has shown measurable improvement over previous generations, and where the practical applications of AI are most visible to developers and users alike.

Garlic will undergo extensive post-training with specialized datasets before launch, along with safety testing and evaluation protocols. The model will not be released as soon as it shows promise but will be refined and validated before deployment.

The competitive dynamic this creates is asymmetrical because Google has already published its Gemini 3 capabilities and the world now knows what it can do, whereas OpenAI is building in secrecy and will have the advantage of benchmarking against known standards.