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

Source: TechCrunch

Image Credits: Jagmeet Singh / TechCrunch

Release Overview

On Thursday, Google released its newest version of Gemini Pro, the powerful LLM now called Gemini 3.1. The model is currently available as a preview, with a general release slated for the near future.

Gemini 3.1 Pro appears to be a significant step up from its predecessor, Gemini 3, which was already regarded as a highly capable AI tool when it launched in November 2025.

Benchmark Performance

Google shared results from independent benchmarks—most notably one called Humanity’s Last Exam—showing Gemini 3.1 Pro outperforming its previous version by a wide margin.

The model was also highlighted by Brendan Foody, CEO of AI startup Mercor. Foody’s benchmarking system, APEX, measures how well new AI models handle real professional tasks. In a social‑media post, he noted:

“Gemini 3.1 Pro is now at the top of the APEX‑Agents leaderboard,”
— Brendan Foody, X post

He added that the results demonstrate “how quickly agents are improving at real knowledge work.”

Industry Context

The release arrives as the AI model wars intensify, with tech companies racing to launch increasingly powerful LLMs geared toward agentic work and multi‑step reasoning. Major players such as OpenAI and Anthropic have also recently introduced new models, further heating up the competition.