Google on Feb. 19 released Gemini 3.1 Pro in preview across consumer, developer and enterprise channels.
Developers and enterprises can access the model through the Gemini API via AI Studio, Antigravity, Vertex AI, Gemini Enterprise, Gemini CLI and Android Studio.
On the consumer side, Gemini 3.1 Pro is rolling out in the Gemini app with higher limits for Google AI Pro and Ultra Plans and is available in NotebookLM exclusively for Pro and Ultra users.
Vertex AI lists gemini-3.1-pro-preview as “Public preview” with a release date of Feb. 19, 2026, and a knowledge cutoff of January 2025.
What Google says changed
Google said Gemini 3.1 Pro reached a “verified score” of 77.1% on ARC-AGI-2 and described ARC-AGI-2 as a test of solving entirely new logic patterns, adding that the result is more than double Gemini 3 Pro’s score.
In its model card, Google listed additional results “as of February 2026,” including 94.3% on GPQA Diamond and 80.6% on SWE-Bench Verified, alongside comparative scores for other named models.
ARC Prize publishes ARC-AGI-2 results on its leaderboard and notes that entries marked “preview” are unofficial and may reflect incomplete testing, while the ARC Prize Verified program indicates verified results.
Google’s developer documentation says Gemini 3.1 Pro Preview is optimized for software engineering behavior and “agentic workflows” that require precise tool usage and reliable multistep execution.
Pricing and limits
Google’s Gemini API pricing page lists Gemini 3.1 Pro Preview at $2.00 per 1 million input tokens and $12.00 per 1 million output tokens for prompts up to 200,000 tokens, with higher rates for prompts above 200,000 tokens.
On the model documentation page, Google lists an input token limit of 1,048,576 and an output token limit of 65,536 for gemini-3.1-pro-preview. Google’s Gemini API documentation also presents the model’s context window and notes the January 2025 knowledge cutoff.
Google flags the endpoint as “Preview,” and its pricing page notes preview models may change before becoming stable and may have more restrictive rate limits.
Competitive context in published evals
Google’s model card includes APEX-Agents results and reports Gemini 3.1 Pro’s score at 33.5% versus 18.4% for Gemini 3 Pro in the same table.
The 33.5% result also exceeds the 24.0% set by Gemini 3 Flash when the benchmark was first introduced in January 2026. It was the highest score reported in the benchmark’s initial paper.
APEX-Agents is a benchmark for long-horizon, cross-application tasks created by professionals in investment banking, management consulting and corporate law, and the authors report Pass@1 results across eight agents.
For pricing comparisons, Anthropic’s Claude Opus 4.6 pricing page lists starting API prices of $5 per 1 million input tokens and $25 per 1 million output tokens. However, requests exceeding 200,000 input tokens incur higher rates of $10 and $37.50, respectively.
Platform note
Google lists Antigravity among access points for Gemini 3.1 Pro preview, and separately describes Antigravity as an agentic development platform for developers building and managing coding agents.