Race to survive: OpenAI throws all resources behind ‘Garlic’ after ChatGPT slump

OpenAI is accelerating development of a new large language model internally codenamed Garlic, as the company seeks to strengthen its position against Google’s Gemini 3 and Anthropic’s Opus 4.5.

According to reports in The Information and The Wall Street Journal, Chief Research Officer Mark Chen briefed employees privately, saying the model is already producing strong results in coding and reasoning tests. Garlic is expected to debut as a mid-cycle upgrade — potentially branded GPT-5.2 or GPT-5.5 — with an anticipated release in early 2026.

The fast-tracking follows what CEO Sam Altman has reportedly labelled a “code red” phase for OpenAI. With competitors advancing quickly, Altman has instructed teams to concentrate on improvements to ChatGPT’s speed, accuracy, reliability and personalisation. Several experimental projects — including prototype advertising tools, health-sector AI initiatives, shopping agents and the Pulse assistant concept — have been paused so that resources can be redirected to core model development. Daily progress check-ins have also been introduced.

Garlic builds on the earlier Shallotpeat programme, incorporating advances in pretraining efficiency and new techniques that transfer capabilities from larger models into smaller, more efficient systems. Early internal evaluations suggest Garlic is performing favourably when compared with Google’s Gemini 3 and Anthropic’s Claude Opus 4.5. Both rival models, launched in November, have been strong performers in multimodal tasks, agentic workflows and code generation.

The push comes amid shifting user dynamics in the AI market. After Gemini 3’s launch, ChatGPT’s daily active users reportedly fell by around 6 percent, from 203 million to 191 million, although weekly usage remained about 800 million in October. Google has reported rapid adoption of Gemini, reaching 650 million monthly users by the same period. Anthropic’s Opus line has also continued to gain attention, particularly among users working with complex coding and workflow automation.

Industry reactions have added to the competitive pressure. Salesforce CEO Marc Benioff said he had switched to Gemini 3 in his workflow, describing Google’s model as extremely strong. Meanwhile, OpenAI is pursuing a valuation of up to $500 billion, though analysts have noted that Google’s infrastructure scale provides a strategic advantage.

OpenAI is also preparing a new reasoning-focused upgrade, expected next week, aimed directly at outperforming Gemini 3 in complex tasks. Garlic, if released on schedule, is projected to deliver more substantial gains in reasoning, stability and code generation in the first half of 2026.