Google Releases Gemini 3.1 Pro, Claiming More Than Double Reasoning Performance

Google announced Gemini 3.1 Pro on February 19, 2026, through its official blog, positioning the model as an upgrade focused on handling complex tasks that require advanced reasoning.

Credit: Google

Google stated that the model improves core intelligence for synthesizing data, explaining intricate topics, and solving problems beyond basic responses.

Gemini 3.1 Pro began rolling out the same day across multiple platforms, including the Gemini app and NotebookLM for consumers, the Gemini API and AI Studio for developers, and Vertex AI and Gemini Enterprise for business users.

The update maintains the existing API pricing at $2 per million input tokens and $12 per million output tokens, along with a one-million-token input context window and a 64,000-token output limit.

The release emphasizes enhanced performance on benchmarks that test logical reasoning and problem-solving. On ARC-AGI-2, which measures a model's capacity to address novel logic patterns not seen in training data, Gemini 3.1 Pro achieved a verified score of 77.1%.

Gemini 3.1 Pro Benchmark Results. Credit: Google

"This is more than double the reasoning performance of 3 Pro," the company stated in the announcement.

Google highlighted the model's capabilities through specific demonstrations, such as generating website-ready animated SVGs from text prompts, where the output uses pure code for scalability and small file sizes compared to pixel-based videos.

In another example, the model configured a public telemetry stream to build a live aerospace dashboard visualizing the International Space Station's orbit, demonstrating its ability to integrate complex APIs into user-friendly interfaces.

Gemini 3.1 Pro also supports interactive applications, as shown in a demonstration where it coded a 3D simulation of a starling murmuration.

The output included user controls via hand-tracking and a generative audio score that adjusts based on the simulated birds' movements, aimed at aiding researchers and designers in prototyping multisensory experiences.

For creative tasks, the model translated themes from Emily Brontë's "Wuthering Heights" into a functional modern website, reasoning through the novel's tone to design an interface that reflects the story's protagonist and atmosphere.

"Gemini 3.1 Pro is here to help you tackle complex tasks," the company wrote, noting the upgrade's rollout in preview form to gather feedback and refine agentic workflows.

The model integrates into GitHub Copilot for coding assistance, with early tests showing high precision in edit-and-test loops and fewer tool calls needed for resolutions.

Google made Gemini 3.1 Pro available immediately in public preview for Copilot Pro, Pro+, Business, and Enterprise users.