New Google Gemini 3.1 Pro give 77% increase in efficiency

Google announced the launch of Gemini 3.1 Pro, an updated version with enhanced core intelligence designed to solve complex problems. This release follows a major update to Gemini 3 Deep Think and is aimed at expanding the practical application of advanced machine thinking in science, research, and engineering, as well as in everyday work.

Mid-Cycle Update and New Version Logic

Last November, the company introduced Gemini 3 Pro as a preview, followed a month later by a lightweight version, Gemini 3 Flash. This is the first time the .1 index has been used to denote a mid-cycle update. Previously, the .5 index scheme was used—specifically, version 2.5 Pro was announced in March and updated in May before the Google I/O conference.

According to the developers, version 3.1 Pro features updated core intelligence, which previously debuted in Deep Think mode. This specialized mode is designed for problems without clearly defined boundaries or a single correct answer, as well as data containing gaps or noise. Its development was carried out with input from the scientific community to adapt the model to modern challenges in research and engineering. The approach is built on a combination of fundamental knowledge and applied performance, with an emphasis on practical results.

Benchmark Performance and Application Scenarios

Update 3.1 Pro is positioned as a step forward in basic reasoning. On the ARC-AGI-2 benchmark, which evaluates the ability to handle previously unknown logical patterns, the model demonstrated a confirmed score of 77.1 percent. This score is more than double the result of the Gemini 3 Pro. Previously, Deep Think mode scored 84.6 percent on the same benchmark, a result confirmed by the ARC Prize Foundation. Without the use of auxiliary tools, the model scored 48.4 percent on Humanity’s Last Exam, achieved an Elo rating of 3455 on the Codeforces platform, and demonstrated gold medal-winning performance at the 2025 International Mathematical Olympiad.

In practical applications, the model is focused on problems where a simple answer is insufficient. The stated use cases include visual explanation of complex topics, integrating disparate data into a coherent structure, and supporting creative projects. The developers emphasize that the emphasis is on comprehensive problem solving rather than formal text generation.

Access for developers, businesses, and users

Since February 19, version 3.1 Pro has been deployed in developer preview via the Gemini API in Google AI Studio, Gemini CLI, Google Antigravity, and Android. Studio. The model is now available to enterprise clients through Vertex AI and Gemini Enterprise. Consumers can access it through the Gemini app and NotebookLM.

The mobile app has increased limits for Google AI Pro and Ultra subscribers. In NotebookLM, access to the new version is available exclusively to users of these plans. The company explains that the preview status is due to the need for additional validation of updates and further refinement, including the development of agent-based workflows, before the final version of the model is generally available.