OpenAI Develops Covert Garlic AI Model to Compete with Google Gemini 3 and Opus 4.5 in Coding and Reasoning Challenges

The race for AI supremacy intensifies as Microsoft-backed OpenAI develops a large language model named Garlic to compete with Google’s Gemini 3 and Anthropic’s Opus 4.5. This advanced model aims to enhance reasoning and coding capabilities, possibly launching as GPT-5.2 or GPT-5.5 by early 2026. OpenAI’s Chief Research Officer, Mark Chen, reports that Garlic has shown strong performance across various benchmarks, suggesting it might outperform existing models internally evaluated. CEO Sam Altman has initiated a “code red” to fast-track Garlic’s release to reclaim leadership in the AI market. Building on lessons from the Shallotpeat model, Garlic boasts improved pretraining efficiency, enabling a smaller model to achieve GPT-4.5-level performance more cost-effectively. With plans for rigorous post-training and safety testing, Garlic could significantly reshape the AI landscape if its early success is confirmed—putting OpenAI back in the competitive forefront against industry rivals.

Source link