gemini3.us
Released February 19, 2026

Gemini 3.1 Pro

Google's most capable AI model — with major advances in reasoning, coding, and agentic intelligence

2× the reasoning of Gemini 3 Pro • #1 on 12 of 18 benchmarks • Native SVG & 3D code rendering

Explore Benchmarks
77.1%
ARC-AGI-2
94.3%
GPQA Diamond
80.6%
SWE-Bench Verified
2887
LiveCodeBench Elo

What's New in Gemini 3.1 Pro

Every dimension reimagined — reasoning, coding, multimodal understanding, and speed

2× Abstract Reasoning

Scores 77.1% on ARC-AGI-2 — more than double Gemini 3 Pro. Solves novel visual-logic puzzles requiring multi-step abstraction that previous models could not.

1M Token Context, 65K Output

Process entire codebases, 8.4 hours of audio, 900-page PDFs, or 1 hour of video in a single prompt. Output expanded to 65,536 tokens — no more truncation.

Elite Software Engineering

80.6% SWE-Bench Verified pass rate. LiveCodeBench Pro Elo of 2887 — outperforming GPT-5.2. Handles real-world software issues autonomously.

Native SVG & 3D Code Rendering

Generates, animates, and visually renders SVG graphics and 3D code directly from natural language descriptions — a capability not commonly found in other models.

Long-Horizon Agentic Workflows

Scores 33.5% on APEX-Agents, 69.2% on MCP Atlas for tool coordination, and 85.9% on BrowseComp for autonomous web research. Strong performance across agentic benchmarks.

Three-Tier Thinking System

Choose between Low, Medium, or High compute modes. Balance output speed vs. reasoning depth for every task — from quick answers to complex problem solving.

Gemini 3.1 Pro Benchmark Performance

#1 on 12 of 18 tracked benchmarks across reasoning, coding, multimodal, and agentic tasks

Abstract Reasoning

ARC-AGI-2
2× Gemini 3 Pro, beats Claude Opus 4.6
77.1%
GPQA Diamond
PhD-level science & research
94.3%
SciCode
Scientific problem solving
59%
APEX-Agents
Autonomous agent tasks
33.5%

Software Engineering

SWE-Bench Verified
Real-world software issue resolution
80.6%
SWE-Bench Pro
Advanced engineering tasks
54.2%
LiveCodeBench Pro Elo
#1, surpasses GPT-5.2 and Gemini 3 Pro
2887
Terminal-Bench 2.0
File system navigation & dependency management
68.5%

Multimodal Understanding

Audio Input
Up to 8.4 hours per prompt
8.4h
Images per Prompt
Up to 900 individual images
900
Video Input
Up to 1 hour of video content
1h
Context Window
1,048,576 tokens input
1M

Agentic Capabilities

MCP Atlas
Tool coordination accuracy
69.2%
BrowseComp
Autonomous web research
85.9%
Overall Intelligence Index
Artificial Analysis rating (57.0)
57.0
Coding Index
Artificial Analysis rating (55.5)
55.5

About Gemini 3.1 Pro

Gemini 3.1 Pro is Google DeepMind's most advanced AI model, released on February 19, 2026. Built on a Transformer-based Mixture-of-Experts architecture atop Gemini 3 Pro, it delivers breakthrough improvements across every metric at the same price point: $2 per 1M input tokens and $12 per 1M output tokens.

Important Notice: Gemini3.us is an independent enthusiast community and developer platform. We are not affiliated with, endorsed by, or officially connected to Google LLC. We provide paid access to Google's official Gemini API services to support our infrastructure and operations.

Experience Gemini 3.1 Pro Today

Try Gemini 3.1 Pro through our platform