Google launches Gemini 3.1 Pro with advanced reasoning abilities: Here’s how to start using

Google has launched its new Gemini 3.1 Pro model which will be taking over as the default model on the Gemini app and NotebookLM. The company claims that its new model is designed for handling complex problem-solving and advanced reasoning-related tasks.

Alphabet CEO Sundar Pichai, who is in India for the AI Impact Summit, wrote in a post on X (formerly Twitter), “With a more capable baseline, it’s great for super complex tasks like visualizing difficult concepts, synthesizing data into a single view, or bringing creative projects to life.”

Meanwhile, Google Labs Vice President Josh Woodward, in a separate post, said that the new model is “great at agentic tasks, intricate coding, and data synthesis projects. You should see fewer errors, better logic, and surprisingly good SVGs.”

How does Gemini 3.1 Pro compare to other models?

Gemini 3.1 Pro has a score of 77.1% on ARC-AGI 2, a benchmark which evaluates models’ ability to solve entirely new logic patterns. Notably, this is over double the reasoning performance of Gemini 3 Pro and much higher than the 52.9% score of GPT-5.2 and the 68.8% score of Claude Opus 4.6.

On the highly coveted Humanity’s Last Exam benchmark, Gemini 3.1 Pro also leads with a score of 44.4%, compared to 40.0% for Opus 4.6 and 34.5% for GPT-5.2. However, when tools like search and code were allowed, Claude Opus 4.6 took a slight lead at 53.1% versus Gemini’s 51.4%.

Gemini 3.1 Pro is still slightly below Opus 4.6 on the SWE-Bench Verified benchmark, which evaluates performance on agentic coding. Opus 4.6 had a score of 80.8% compared to an 80.6% score for Gemini 3.1 Pro and an 80% score for GPT-5.2.

On the APEX-Agents benchmark, Gemini 3.1 Pro led with a score of 33.5%, ahead of Opus 4.6 (29.8%) and GPT-5.2 (23.0%).

Similarly, on Terminal-Bench 2.0, Gemini had a score of 68.5%, ahead of Opus 4.6 (65.4%) and GPT-5.2 (54.0%).

Gemini 3.1 Pro availability:

Google says that Gemini 3.1 Pro is rolling out to consumers via the Gemini app and NotebookLM. The model is available for free in Gemini, with higher limits for Pro and Ultra users. Meanwhile, the NotebookLM rollout is available only to Pro and Ultra users.

The model is also in preview for developers via the Gemini API in Google AI Studio, Gemini CLI, the Google Antigravity agentic development platform, and Android Studio. Meanwhile, enterprise users can access the model through Vertex AI and Gemini Enterprise.