OpenAI’s Comeback with “Garlic”
According to The Information, OpenAI is developing a new large language model codenamed “Garlic” to counter Google’s recent gains with Gemini 3.
Background of the “Code Red” Declaration
Last week, after Google released Gemini 3 to strong praise from the AI community, OpenAI CEO Sam Altman declared a “code red” internally. He directed the team to freeze all non-essential projects—including shopping agents, health tools, advertising agents, and the new personal assistant Pulse—and focus everything on improving ChatGPT.
Garlic’s Performance
OpenAI’s Chief Research Officer Mark Chen informed colleagues that Garlic has performed well in company evaluations:
- vs. Gemini 3: Competitive performance in coding and reasoning tasks
- vs. Opus 4.5: Similarly competitive results
- Pretraining improvements: Fixed issues found in previous model “Shallotpeat”
Technical Breakthrough
In developing Garlic, OpenAI has improved pretraining techniques—the initial phase where the model learns from a massive dataset. This enables smaller models to be infused with the same amount of knowledge previously reserved for larger models.
This technical advancement means:
- Cost Reduction: Achieving comparable performance with smaller models
- Faster Development: More efficient training processes
- Better Scalability: Easier deployment of models at various sizes
Steps Before Release
Garlic still requires:
- Post-training with curated data
- Further testing and validation
- Comprehensive safety evaluations
Intensifying AI Competition
This move demonstrates the intensifying competition among the top three AI companies (OpenAI, Google, Anthropic). With Salesforce CEO Marc Benioff tweeting “I’m not going back” after trying Gemini 3, OpenAI has reason for concern.
While the release timeline for Garlic hasn’t been announced, given OpenAI’s company-wide focus on development, an announcement may come soon.