OpenAI is accelerating the development of its next-generation large language model, codenamed "Garlic," in response to competitive pressures from Alphabet's recent AI breakthroughs, aiming to reclaim its technological leadership.
On December 2, sources familiar with the matter revealed that OpenAI's Chief Research Officer, Mark Chen, stated in an internal communication last week that Garlic has performed well in internal evaluations, surpassing Alphabet's Gemini 3 and Anthropic's Opus 4.5 in key tasks such as programming and reasoning. The company plans to release related versions soon, suggesting that GPT-5.2 or GPT-5.5 could debut by early 2026.
This accelerated push comes as Alphabet gains market attention with Gemini 3. Earlier reports indicated that OpenAI CEO Sam Altman initiated a "red alert" to enhance ChatGPT's capabilities comprehensively. As a first step, OpenAI is set to launch a reasoning model that is "significantly ahead" of Gemini 3 in internal evaluations.
**Technical Breakthroughs in Garlic Compared to Previous Models** Garlic is a standalone model distinct from Shallotpeat, another large language model under development at OpenAI. Altman informed employees in October that Shallotpeat would help OpenAI challenge Gemini 3. Garlic incorporates error fixes from the pre-training phase of Shallotpeat's development.
This is critical, as Alphabet recently claimed breakthroughs in pre-training for Gemini 3, which OpenAI's leadership has acknowledged. Chen noted that OpenAI resolved key pre-training issues during Garlic's development, including optimizations for GPT-4.5—a model touted as "best-in-class" and larger in scale but rarely mentioned since its February release.
Chen emphasized that these improvements allow OpenAI to embed comparable knowledge into smaller models, a feat previously achievable only with larger, costlier, and time-consuming models. Leveraging lessons from Garlic, OpenAI has begun developing even larger and more powerful models.
**Multiple Tests Required Before Release** Before its official launch, Garlic must undergo additional steps, including post-training (exposure to curated data for specialized domains like medicine or law) and further safety evaluations.
It remains unclear how developers will assess Garlic in real-world applications. As history shows, strong internal performance does not always translate to consistent results in practical scenarios like programming or data analysis—GPT-5's market reception serves as a case in point.
Amid the current competitive challenges, Chen and OpenAI's leadership also play a role in boosting internal morale, fostering confidence within the team.