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

On Thursday, Google introduced Gemini 3.1 Pro, the latest iteration of its advanced language model, now available for preview with a general release on the horizon. This new model is anticipated to be among the strongest in the field of large language models (LLMs). Observers are noting that Gemini 3.1 Pro represents a significant enhancement over its predecessor, Gemini 3, which was already regarded as a formidable AI tool upon its launch last November. Accompanying the announcement, Google released independent benchmark results, including from a test called Humanity’s Last Exam, which highlighted the model's superior performance compared to earlier versions. Brendan Foody, CEO of the AI startup Mercor, praised Gemini 3.1 Pro for its achievements in his benchmarking system, APEX, which evaluates how effectively AI models can tackle real-world professional tasks. "Gemini 3.1 Pro is now leading the APEX-Agents leaderboard," Foody remarked in a social media update, emphasizing that the model's remarkable outcomes illustrate the rapid advancements agents are making in applied knowledge work. The launch of Gemini 3.1 Pro arrives amidst intensifying competition in the AI landscape, as major tech players like OpenAI and Anthropic are also rolling out their latest models, all designed to push the boundaries of agentic work and complex reasoning capabilities.