Google's Gemini 3.1 Pro Doubles Reasoning Performance in Benchmark Tests

Google has released Gemini 3.1 Pro, featuring advanced reasoning capabilities designed for complex, multi-step tasks.

Benchmark Performance

The new model achieved impressive results:

• 77.1% score on ARC-AGI-2 benchmark
• Roughly double the performance of Gemini 3 Pro
• Record benchmark scores in reasoning tasks

New Capabilities

Gemini 3.1 Pro can now:

• Generate code-based animated SVGs
• Build complex applications like city planning simulators
• Synthesize large datasets into actionable insights
• Handle sophisticated multi-step reasoning chains

Phone Automation Coming

Google is also previewing new Gemini automation features launching March 11th:

• Beta for Pixel 10 and Galaxy S26 devices
• Handles routine tasks like ordering food and booking rides
• Operates within a secure sandbox for privacy
• Integration with ride-sharing, grocery, and food apps

Updated Pricing

Google’s AI subscription tiers have been restructured:

• Google AI Pro: $19.99/month (formerly Google One AI Premium)
• Google AI Ultra: Higher tier with additional features