Google on Wednesday announced the rollout of Gemini 3.1 Pro, a major update to its flagship large language model, aimed at improving reasoning accuracy and performance on complex, multi-step tasks.
The new model is being released in preview across consumer and developer-facing platforms, including the Gemini app, NotebookLM, and Google’s developer tools such as Google AI Studio and Vertex AI. The company said the update reflects a targeted intelligence upgrade rather than a broad expansion of features.
A focused upgrade to core intelligence
Gemini 3.1 Pro marks Google’s first-ever “.1” version increment, breaking from its earlier practice of using “.5” releases for mid-cycle updates. According to Google, the change in naming signals a refinement of core reasoning capabilities rather than a cosmetic or feature-heavy refresh.
Also Read | India built what no nation has: French President Macron praises UPI at AI Impact Summit
At the heart of the update is the advanced reasoning engine first introduced in Gemini 3 Deep Think earlier this month. With 3.1 Pro, those capabilities are now being made available to a wider audience.
Google said the model is designed for scenarios where simple question-and-answer responses are insufficient. The system is built to handle tasks such as synthesising large datasets, solving layered scientific and technical problems, and generating complex outputs like animated SVG files directly from text prompts. These SVG outputs are code-based rather than pixel-rendered, allowing them to remain sharp at any scale while keeping file sizes small.
Benchmark gains against rival models
Google pointed out that there were major improvements in various industry benchmarks. Gemini 3.1 Pro showed a 77.1% score in the ARC-AGI-2 test, which is intended to assess abstract reasoning skills on unfamiliar logic patterns, more than twice the score of Gemini 3 Pro.
The model also showed a 94.3% result in the GPQA Diamond benchmark, which is intended to assess advanced scientific knowledge, and 80.6% in the SWE-Bench Verified benchmark, a benchmark focused on agentic coding tasks. In BrowseComp, which is intended to assess agent-based web search, the model showed an 85.9% result.
Also Read | Think you can spot a scam? AI-powered deepfakes and phishing are making it harder than ever
In competitive coding tests, Gemini 3.1 Pro demonstrated an Elo rating of 2887 in the LiveCodeBench Pro test, making it superior to similar models from competing companies such as OpenAI and Anthropic, according to figures cited by Google.
Access and rollout plans
Gemini 3.1 Pro is currently available in preview as Google continues to validate its performance, particularly for agent-based workflows. Consumers can access the model through the Gemini app and NotebookLM, with higher usage limits offered to Google AI Pro and Ultra subscribers.
Developers can use the model via the Gemini API through platforms including Google AI Studio, Vertex AI, Gemini CLI, Antigravity, Gemini Enterprise, and Android Studio. Google said general availability is expected in the near future.
