OpenAI pushes development of ‘Garlic’ AI model to challenge Google’s Gemini 3

OpenAI is intensifying its race against Google as it develops a new high-performance AI model codenamed Garlic, a project reported by The Information as a direct response to the rapid success of Google’s Gemini 3. The model, which is being positioned to rival Gemini 3’s dominance on the LMArena AI leaderboard, reportedly prompted CEO Sam Altman to declare a code red internally, signalling an urgent push to advance ChatGPT and regain competitive leadership in the AI space.

According to the report, OpenAI’s Chief Research Officer Mark Chen informed employees that Garlic has performed strongly in internal benchmarking, outscoring Google’s Gemini 3 and Anthropic’s Opus 4.5 in both coding and reasoning tasks. These areas are seen as critical battlegrounds, with Gemini 3 recognised for its reasoning abilities and Opus 4.5 widely viewed as the industry’s strongest coding-focused model.

Chen stated that while no firm launch timeline has been set, the ambition is to release Garlic as soon as possible. Given the pace of development and competitive pressure, the model could enter the market in early 2026. He also informed staff that the breakthroughs achieved while building Garlic have already contributed to the groundwork for OpenAI’s next major model, highlighting improvements in pretraining methods that have enabled the creation of more efficient and smaller systems.

The report further noted that Garlic is distinct from another model Altman previously mentioned to employees in October, known internally as Shallotpeat. As the competitive landscape continues to shift, OpenAI’s internal messaging suggests a strong drive to maintain relevance, reduce model-building costs, and push forward with innovations that match or exceed the capabilities of its leading rivals.