OpenAI is working on a secret garlic AI model to challenge Google Gemini 3 and Opus 4.5 in coding and reasoning
OpenAI has been quietly developing Garlic, a new AI model to compete with Google’s Gemini 3. According to a new report, the upcoming model has shown “strong performance” in several benchmarks, including logic and programming.
The race to build the most powerful artificial intelligence system has entered a new phase, with Microsoft-backed OpenAI quietly developing a larger language model called Garlic. According to a report by The Information, the model is being designed specifically to rival Google’s Gemini 3 and Anthropic’s Opus 4.5 in advanced logic and coding capabilities. Early internal testing suggests the garlic is performing strongly and may emerge as GPT-5.2 or GPT-5.5 as early as next year.
The Garlic project comes amid increasing competition following Google’s success with Gemini 3. Mark Chen, OpenAI’s chief research officer, reportedly told colleagues that Garlic has shown “strong performance” in a number of benchmarks, including logic and programming, where Google and Anthropic currently hold the lead.
CEO Sam Altman has reportedly announced a “code red” inside the company to improve ChatGPT and regain OpenAI’s lead in the AI race. He told staff that OpenAI’s new reasoning model was already “ahead” of Gemini 3 in its internal evaluation. Although OpenAI has not commented publicly, insiders say the company is fast-tracking the release of Garlic, aiming for an early 2026 launch.
OpenAI’s new model for smart AI
Garlic reportedly took lessons from ShallotPete, a previously in-house model, which Altman mentioned to employees in October. While Shallotpeat was designed to challenge Gemini 3, Garlic incorporates bug fixes and refinements from that project, particularly in the pretraining phase. This step teaches a model to recognize relationships in huge datasets pulled from the Internet.
According to Chen, garlic represents a leap forward in pre-training efficiency. He told colleagues that the team had managed to “put the same amount of knowledge into a smaller model” that previously required much larger models. This advancement means garlic can deliver GPT-4.5-level performance at a lower cost and faster rate.
This success comes as Google is announcing similar improvements with Gemini 3’s training process. OpenAI’s advancements with Garlic could balance out that advantage and potentially provide the company with a more efficient path for future upgrades.
Chen said Garlic has already surpassed OpenAI’s “previous best” pretraining results and solved major technical hurdles affecting GPT-4.5, which launched earlier this year. With these improvements, OpenAI is confident that it can now develop smaller but more capable models without increasing training costs.
Before the launch of Garlic, it will undergo security testing and evaluation as well as training with special datasets. Sources also claim that the success of Garlic has allowed OpenAI to begin working on even more advanced successor models based on lessons learned during its development.
If Garlic performs as strongly as early reports suggest, it could signal a major shift in the balance of power between the AI giants. As OpenAI, Google and Anthropic strive for dominance, the competition over next-generation reasoning models is heating up, and Garlic may be OpenAI’s boldest move yet.
