Gemini 3.1 Pro: Google's Greatest Model Ever — Full Review & Testing

Google is back with Gemini 3.1 Pro — a smarter model designed for complex tasks where a simple answer isn’t enough. This is a full hands-on review covering benchmarks, coding, simulations, and real-world testing.

Benchmark Performance

ARC-AGI-2: Gemini 3.1 Pro scores 77.1% — more than 2x the performance of Gemini 3 Pro. This is a massive leap in core reasoning.

Other benchmarks:

• State-of-the-art in multiple areas
• Strong SWE-bench verified score, though slightly behind Opus 4.6 by a marginal amount
• Excellent LiveCodeBench results compared to other frontier models

Pricing & Access:

• 1 million token context window
• $2 per 1M input tokens,$12 per 1M output tokens
• Available via Google AI Studio, Gemini App, Vertex AI, API, OpenRouter, and Kilo (which offers $25 free credits)

Frontend & Vibe Coding

Minecraft Clone

Using the leaked 3.1 version, the model generated a fully functional Minecraft clone with:

• Block breaking and pickup mechanics
• Underground cave system generation
• Terrain generation with proper layering

Browser-Based macOS

Using the Kilo CLI, Gemini 3.1 Pro built a browser-based OS mimicking macOS:

• Clean, minimalistic toolbar
• SVG icons generated for each app (Safari, Notes, Music, Terminal, Activity Monitor, Calculator, Settings, Email)
• Working apps with proper functionality
• Login screen with notification popups
• Best generation seen from any model for this benchmark

Landing Pages

When given an image reference, the model accurately coded out landing pages with:

• Perfect, smooth animations
• Proper typography
• Dynamic functionality throughout
• Significantly more robust than generations from previous models using the same prompts

SVG Generation — Best in Class

Gemini 3.1 Pro is arguably the best model for SVG generation:

Animated Butterfly:

• Best animated butterfly seen — functional wing movements
• Better than Opus 4.6’s non-functional output
• Unique design with proper motion mimicry

Photorealistic Butterfly (reprompted):

• Improved element detail, though animations still not fully realistic
• Opus 4.6 produced comparable quality on the reprompt

iOS App Recreation:

• Given a screenshot of an iOS app, the model coded out every element one by one
• Added functionality and SVG assets for all components

Goldfish Aquarium:

• Realistic goldfish in SVG within an aquarium
• Multiple animated elements: seagrass movement, fish backbone animation, bubble inflation

Engineering & Simulation

Double Wishbone Suspension System

The model fully coded a double wishbone suspension system with:

• Independent suspension geometry
• Dynamic coilover shocks
• Vented performance disc brakes
• Steering simulation integrated with brake system
• Complex reasoning applied to accurately showcase real mechanical behavior

City Planner Application

One of the most impressive demos:

• Analyzed terrain and designed infrastructure
• Laid out roads and districts
• Simulated traffic flow
• Used reasoning about geography, movement, and urban design simultaneously
• Produced a high-quality interactive visualization

360° Product Viewer

An interactive iPhone product viewer with:

• Full 360° rotation and zoom
• Clickable component hotspots (Action Button, Pro Camera System, etc.)
• Color variant switching with animations
• Spatial reasoning accurately representing the device from all angles

Solar System Simulation

A Three.js solar system simulation featuring:

• Accurate orbital mechanics for each planet
• Moon systems rendered correctly
• Proper distance scaling from the sun
• Smooth simulation of all planetary bodies

Strengths & Weaknesses

Strengths:

• Massive reasoning leap (2x+ over Gemini 3 Pro)
• Best-in-class SVG generation
• Excellent frontend/vibe coding
• Strong simulation and engineering capabilities
• Pushes the Pareto frontier of performance and efficiency

Weaknesses:

• Still somewhat “lazy” like the previous checkpoint — sometimes cuts corners
• Occasional hallucinations in agent capabilities
• Still not surpassing Opus 4.6 in agent tasks, though gap is narrowing
• Not a new model family (still the 3.x series)

Verdict

Gemini 3.1 Pro represents a major leap in reasoning and complex task execution. It pushes AI closer to real-world engineering and system-level thinking. It’s particularly strong for frontend development, simulations, and multi-step problem solving. While still not perfect, the gap to the top models is closing fast — and this is just a 3.1 release, with 3.5 Pro and the 4.0 series expected later this year.