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

On Thursday, Google introduced the newest version of Gemini Pro, its high-end large language model. The company said the Gemini 3.1 Pro model is now available in preview, with a broader general release coming soon.

Google’s latest model is shaping up to be one of the most capable LLMs yet. Observers have said Gemini 3.1 Pro looks like a notable leap over its predecessor, Gemini 3, which, when it launched in November, was already viewed as a highly advanced and competitive AI system.

Alongside the announcement, Google also published results from independent benchmarks — including one known as Humanity’s Last Exam — showing the new model performing materially better than the previous version.

Gemini 3.1 Pro also drew praise from Brendan Foody, the CEO of AI startup Mercor, whose APEX benchmarking system is built to evaluate how well modern AI models handle real professional tasks. “Gemini 3.1 Pro is now at the top of the APEX-Agents leaderboard,” Foody wrote in a social media post, adding that the strong showing highlights “how quickly agents are improving at real knowledge work.”

The release arrives as competition in the AI model race intensifies with major tech firms regularly shipping more powerful LLMs geared toward agentic workflows and multi-step reasoning. Other leading players — including OpenAI and Anthropic — have also recently launched new models.