Gemini 3 had blown away competition on benchmarks when it had released last month, but OpenAI now says that they already have models internally that perform at the same level.
OpenAI’s Chief Research Officer Mark Chen has revealed that the company possesses unreleased AI models matching Google’s latest Gemini 3 performance, signaling that the competitive race in frontier AI development remains tighter than public releases might suggest. Chen’s remarks offer a rare glimpse into OpenAI’s internal development pipeline and underscore a broader industry trend: benchmark leadership is increasingly ephemeral, and what companies choose to release publicly may lag significantly behind their internal capabilities.
Speaking about Gemini 3 specifically, Chen acknowledged Google’s achievement while tempering expectations around benchmark supremacy. “It’s a pretty good model,” he said, before adding a crucial caveat: “One thing we do is try to build consensus. The benchmarks only tell you so much. Just looking purely at the benchmarks, we actually felt quite confident.”
Chen then made the big assertion: “We have models internally that perform at the level of Gemini 3, and we’re pretty confident that we’ll release them soon and we can release successor models that are even better.”
However, the OpenAI executive was quick to reiterate his skepticism about over-indexing on benchmark performance. “Again, the benchmarks only tell you so much. I think everyone probes the models in their own way,” he explained, suggesting that real-world performance and user experience may diverge significantly from standardized testing metrics.
Chen’s comments highlight a strategic tension facing all major AI labs: when to release cutting-edge capabilities versus holding them back for refinement, safety testing, or competitive timing. The revelation that OpenAI has Gemini 3-level models “internally” raises questions about the company’s release cadence and what might be coming next. If OpenAI indeed possesses models matching Gemini 3’s capabilities—and has even more powerful successors in development—the company may be preparing a significant product update that could reshape the competitive landscape once again. This pattern of leapfrogging has defined the AI race, with Anthropic’s Claude, Google’s Gemini, and OpenAI’s GPT series trading benchmark leadership over the past year. But given how some companies seem to have fallen by the wayside as well — Meta is no longer at the frontier, for instance — this leapfrogging might not continue forever, and a company or two could end up definitively dominating in the coming years.