OpenAI is cooking up a new Garlic model to fight back Google

For the past three years, the entire tech industry moved at OpenAI's pace. Now, the tables have turned, and OpenAI is the one scrambling to keep up.

OpenAI is accelerating the development of its next flagship language model, codenamed Garlic, with reports indicating it could launch early in the new year. According to The Information, OpenAI chief research officer Mark Chen briefed colleagues on the project last week, and the model is expected to be released as either GPT-5.2 or GPT-5.5.

Inside OpenAI, CEO Sam Altman reportedly declared a "code red" this week, demanding rapid improvements to ChatGPT as rival releases eat into its momentum.

A diet of smaller data and faster timelines

Garlic is reportedly trading blows with Google's Gemini 3 and Anthropic's Opus 4.5 on coding and reasoning benchmarks. But the real twist is its diet: the model was trained on a much smaller dataset than GPT-4.5, signaling an intentional push toward training efficiency and lower operating costs.

It is not the only project in the kitchen. OpenAI has also been working on another model codenamed Shallotpeat since at least October. Garlic reportedly integrates bug fixes identified during Shallotpeat's pretraining, and the two efforts may end up being merged as the launch schedule gets compressed.

At the same time, OpenAI is under immense financial pressure to pay for its massive infrastructure spend. The company rolled out a shopping assistant right before Black Friday and is currently testing conversational ad units in its Android beta under labels like "search ad" and "bazaar content."

Why it matters

We think this is the clearest sign yet that OpenAI has lost its undisputed lead and is now operating in pure defense mode.

Google has erased the narrative that it fumbled generative AI, bouncing back with Nano Banana and Gemini 3 to amass 650 million users, rapidly breathing down ChatGPT's 800 million weekly active user base. Meanwhile, Anthropic has firmly locked down the high-end developer and enterprise crowd, crossing 300,000 business customers and earning a reputation for superior coding chops.

Pivoting Garlic toward smaller datasets is a smart, pragmatic play. Running giant models is brutally expensive, and whoever can deliver frontier reasoning at commodity costs wins the long game. However, if Garlic merely matches Gemini 3 rather than blowing past it, OpenAI risks looking like just another AI vendor. Coupling that with rushed experiments like ads in chat responses could alienate its loyal core just when the competition is at its fiercest.