Google Launches Gemini 3.1 Pro With ARC-AGI-2 Score of 77.1% and Broad Rollout Paths

Announcement Scope: One Model, Multiple Distribution Channels

Google’s 2026-02-19 announcement positions Gemini 3.1 Pro as an upgraded core reasoning model intended for complex tasks. The important operational detail is not only the model update itself, but the breadth of launch channels activated at the same time. Rather than limiting access to a single sandbox, Google describes coordinated rollout tracks for developers, enterprises, and end users.

For developers, the company says 3.1 Pro is available in preview via Gemini API in Google AI Studio, Gemini CLI, Google Antigravity, and Android Studio. For enterprises, it points to Vertex AI and Gemini Enterprise. For consumers, it highlights Gemini app and NotebookLM access. This packaging suggests Google is prioritizing cross-surface consistency and faster feedback loops across product layers.

Published Performance Signal: ARC-AGI-2 at 77.1%

The post states that Gemini 3.1 Pro achieved a verified 77.1% score on ARC-AGI-2 and frames this as more than double the reasoning performance of 3 Pro. As with all benchmark claims, comparability depends on setup and test protocol, but Google is explicitly using a difficult reasoning benchmark as the headline signal for this release.

Google also emphasizes that 3.1 Pro is in preview, not full general availability yet. The stated rationale is to validate updates and continue advancing ambitious agentic workflows before broad GA. In practice, that means capability claims are paired with staged operational risk management.

Why This Matters for AI/IT Teams

• Developer tooling: API + CLI + IDE paths lower friction for rapid prototyping and agent experiments.
• Enterprise adoption: Vertex AI and Gemini Enterprise provide a clearer route for governed deployment.
• End-user exposure: Gemini app and NotebookLM distribution can accelerate real-world usage signals.

The release is therefore both a model update and a platform move. The next checkpoints are preview stability, performance variance on production workloads, and what changes by the time Google declares general availability.

Source: Google Blog