Google introduced the latest iteration of its advanced large language model, Gemini Pro, on Thursday. This new version, labeled 3.1, is currently accessible in a preview phase, with a full public launch expected shortly.

Early assessments suggest Gemini 3.1 Pro represents a substantial advancement over its predecessor, Gemini 3.0, which had already established itself as a highly competent AI system upon its debut last November.

To demonstrate the improvements, Google provided data from independent benchmark tests, including one known as Humanity’s Last Exam. The results indicate the new model performs markedly better than the previous version.

The model also received commendation from Brendan Foody, CEO of the AI startup Mercor. His company’s APEX benchmarking platform evaluates how effectively AI models handle authentic professional assignments. In a social media statement, Foody noted, “Gemini 3.1 Pro is now at the top of the APEX-Agents leaderboard.” He added that its strong performance highlights “how quickly agents are improving at real knowledge work.”

This launch occurs amid intensifying competition in the AI sector, where leading technology firms are consistently unveiling more sophisticated large language models. These new systems are increasingly engineered for agentic tasks and complex, multi-step reasoning. Other prominent organizations, such as OpenAI and Anthropic, have also recently introduced their own updated models.