OpenAI's Quick Move with 'Garlic' to Enhance ChatGPT's Performance: Insights

Overview from ZDNET

In light of Google's recent advancements, the swift ascent of their latest project prompted OpenAI's CEO, Sam Altman, to issue an urgent 'code red.' The objective is to refine ChatGPT to ensure it stands firm in competition, according to a recent report. It has since been revealed that the company is pressing forward with a subsequent model named Garlic.

According to insights from OpenAI's Chief Research Officer, Mark Chen, Garlic has shown promising results in internal assessments, notably in comparison to Gemini 3 and Anthropic's Opus 4.5 when it comes to programming and logical reasoning tasks. This is significant as both Gemini 3 and Opus 4.5, launched in the prior month, have established new benchmarks, with one excelling in logic and the other in programming.

OpenAI has chosen not to provide immediate comments on these findings.

Chen highlighted that during Garlic's development, OpenAI tackled issues associated with the pretraining phase, which is the period when the model starts learning from extensive datasets. This step now focuses on establishing broader interconnections before concentrating on targeted tasks. These adaptations facilitate OpenAI in equipping a compact model with the depth of knowledge that used to be reserved for larger counterparts.

Smaller models bring advantages, particularly for developers, as they often cost less and are simpler to implement. This was pointed out by the French AI research team, Mistral. For OpenAI, a more compact model translates to reduced construction and deployment costs. It is crucial not to confuse Garlic with Shallotpeat—another model brought to light by Altman in October, which also aimed to amend pretraining flaws.

Although specifics about Garlic's release remain vague, with Chen indicating an arrival 'as soon as possible,' the urgency suggests it might debut early next year. The breakthroughs achieved with Garlic are paving the way for OpenAI's future, more advanced models, according to Chen.

Competition for Market Dominance

The intense rivalry between Google and OpenAI seems driven by a pursuit of the same demographic: end-users.

Dario Amodei, Anthropic’s CEO, conveyed to Andrew Ross Sorkin at the DealBook Summit that Anthropic is not participating in the same frantic race as its counterparts, instead concentrating on enterprise solutions. The company's Claude Code tool has remarkably hit $1 billion in revenue in just six months of its launch, underscoring its enterprise focus rather than the consumer market.