Google’s new Gemini Pro model has record benchmark scores—again

In Brief

On Thursday, Google released the newest mentation of Gemini Pro, its almighty LLM. The model, 3.1, is presently disposable arsenic a preview and volition beryllium mostly released soon, the institution said.

Google’s caller exemplary whitethorn beryllium 1 of the astir almighty LLMs yet. Onlookers person noted that Gemini 3.1 Pro appears to beryllium a large measurement up from its predecessor, Gemini 3—which, upon its merchandise successful November, was already considered a highly susceptible AI tool.

On Thursday, Google besides shared statistic from autarkic benchmarks—such arsenic 1 called Humanity’s Last Exam—that showed it performing importantly amended than its erstwhile version.

Gemini 3.1 Pro was besides praised by Brendan Foody, the CEO of AI startup Mercor, whose benchmarking system, APEX, is designed to measurement however good caller AI models execute existent nonrecreational tasks. “Gemini 3.1 Pro is present astatine the apical of the APEX-Agents leaderboard,” Foody said in a societal media post, adding that the model’s awesome results amusement “how rapidly agents are improving astatine existent cognition work.”

The merchandise comes arsenic the AI exemplary wars are heating up, and tech companies proceed to merchandise progressively almighty LLMs designed for agentic enactment and multi-step reasoning. Other large names—including OpenAI and Anthropic—have precocious released caller models arsenic well.

Subscribe for the industry’s biggest tech news