On December 3rd, the Information reported thatOpenAI NEW AI UNDER DEVELOPMENT Large Language ModelIn response to the technical competition of Google Gemini 3, the project is codenamed "Greek."Garlic).
According to the report, the Chief Research Officer of OpenAI, Mark Chen, introduced the new one to the team last week AI ModelsIt is alleged that the model performed better than Google Gemini 3.0 and Anthropic Opus 4.5 in its programming and logical reasoning tasks。
The CEO of OpenAI, Sam Altmann, had previously launched an emergency response mechanism to enhance ChatGPT performance and planned to take the lead in launching a new model that optimized reasoning。
According to the Information, “Garlic” is a different technical route from the previously disclosed “Shallotpeat” and its core breakthrough lies in the optimization of the pre-training phase: a significant reduction in training costs and time by improving the algorithm architecture to achieve the injection of previously large-scale model-based knowledge density into smaller-parametric models. The technical path directly addressed the “pre-training leap” claimed by Google and repaired the key deficiencies of GPT-4.5. OpenAI indicates that, based on this experience, the development of larger models has been initiated。
The current “Garlic” still needs to complete a follow-up process such as security assessment, fine-tuning in specific areas (notes: medical, legal), which is expected to be published in GPT-5.2 or GPT-5.5 by early 2026。