This Thursday, February 19, Google deployed the preview version of the new language model (LLM) of its flagship AI, Gemini Pro 3.1. Successor to Gemini 3 released last November, this new version brings increased performance and even more advanced reasoning.
The final version of Gemini Pro 3.1 will only be released in a few weeks, of course, but we can already get a real glimpse of what it is capable of. The first benchmarks have been released, and give the new model the hands down winner.
Gemini Pro 3.1 shows significantly increased results compared to the competition, and even compared to previous versions of Google’s AI. Two independent benchmarks also gave excellent results to the new model, including Humanity’s Last Exam and APEX-Agents. The first evaluates the intellectual power of the model, its ability to reason, while the second analyzes concrete effectiveness in a work context, the way in which the language model carries out professional tasks (analyzing a contract, writing code, structuring data, etc.).
GPT-5.2 lagging behind
With such results, Gemini demonstrates that it has finally managed to catch up with ChatGPT, or even surpass it. GPT-5.2 failed to compete with Gemini Pro 3.1 in any of the benchmarks carried out by the AI startup Mercor, proof if necessary of the omnipotence of Gemini.
In a context where language models abound and where competition is fierce, three companies are trying to succeed: Google, OpenAI and Anthropic. Each seeks to make its model more autonomous, to offer an agent capable of chaining several actions together without losing track and, above all, to direct their AI towards professional productivity.
🟣 To not miss any news on the WorldOfSoftware, follow us on Google and on our WhatsApp channel. And if you love us, .
