Google DeepMind Releases Gemini 3.1 Pro: #1 on AI Benchmarks, Beats Claude Opus at Lower Cost

Glowing AI neural network brain representing Google Gemini 3.1 Pro intelligence and reasoning capabilities

Google DeepMind has released Gemini 3.1 Pro — a significant update to its most capable everyday AI model. Built on the same core intelligence as Gemini 3 Deep Think, but optimised for practical applications at scale, 3.1 Pro delivers a major leap in reasoning performance and has moved straight to the top of independent AI benchmarks.

Google DeepMind Gemini 3.1 Pro trending on X with reaction from Sundar Pichai and Jeff Dean
Gemini 3.1 Pro trending on X — reactions from Google CEO Sundar Pichai, Jeff Dean, and over 24,000 posts in hours (Source: X/Twitter)

What Is Gemini 3.1 Pro?

Gemini 3.1 Pro is Google's answer to a clear market need: the intelligence of their most powerful research model, made accessible and affordable for real-world use. As Google put it at launch: "It's the same core intelligence that powers Gemini 3 Deep Think, now scaled for your practical applications."

This means developers, enterprises, and everyday users can now access near-Deep Think reasoning without Deep Think pricing. It is rolling out to:

  • Developers via the Gemini API in Google AI Studio (preview)
  • Enterprises via Vertex AI and Gemini Enterprise
  • Everyone through the Gemini app and NotebookLM

The Benchmark Numbers Are Striking

Google CEO Sundar Pichai summarised the jump bluntly: "Hitting 77.1% on ARC-AGI-2, it's a step forward in core reasoning (more than 2x 3 Pro)." Jeff Dean added: "This updated model scores 77.1% on ARC-AGI-2, more than double the reasoning performance of its predecessor."

Sundar Pichai tweet showing Gemini 3.1 Pro benchmark comparison table vs Claude Opus, GPT-5.2 and others
Sundar Pichai's benchmark comparison showing Gemini 3.1 Pro leading across key metrics (Source: @sundarpichai on X)

Here is a breakdown of the key benchmark scores:

  • ARC-AGI-2: 77.1% — up from 31.1% in Gemini 3 Pro, more than doubled
  • GPQA Diamond (scientific reasoning): 94.3%
  • SWE-Bench (software engineering): 80.6%
  • Artificial Analysis Intelligence Index: 57 points — ranked #1, ahead of Claude Opus 4.6 and others

Importantly, independent benchmarking firm Artificial Analysis places Gemini 3.1 Pro at the top of their Intelligence Index — while noting it costs significantly less to run than comparable competitors.

What ARC-AGI-2 Actually Measures

ARC-AGI-2 (Abstraction and Reasoning Corpus) is considered one of the hardest reasoning benchmarks in AI — it tests the ability to solve novel, abstract puzzles that require genuine generalisation rather than pattern memorisation. A jump from 31.1% to 77.1% is not a minor improvement. It represents the kind of qualitative leap in core reasoning that practitioners notice in real tasks like complex coding, multi-step analysis, and scientific problem-solving.

Where It Still Has Ground to Cover

Not all the news is one-sided. According to the X trending summary (sourced from Grok's analysis of thousands of posts), Gemini 3.1 Pro trails some competitors in certain agentic tasks — the kind of long-horizon, multi-step autonomous workflows that are increasingly central to enterprise AI use. One community reply put it well: "Google is winning the logic race, but the agency race is still wide open."

For developers building agents that need to run 24-hour autonomous tasks without intervention, this remains an area to watch. Google is clearly investing here, but 3.1 Pro is not yet the definitive agentic model.

The Competitive Picture

Gemini 3.1 Pro's position at the top of Artificial Analysis' Intelligence Index — at lower cost than rivals — is a strategic signal as much as a technical one. Google is competing not just on raw capability but on the value-to-cost ratio, which matters enormously for API-scale enterprise deployments.

For individual users and developers, the message is simpler: the gap between Google's consumer model and its cutting-edge research model has significantly narrowed with this release.

How to Access Gemini 3.1 Pro

  • Gemini app: Available now, rolling out globally
  • API: Developer preview via Google AI Studio
  • Enterprise: Vertex AI and Gemini Enterprise
  • NotebookLM: Integrated for research and summarisation use cases

The Bottom Line

Gemini 3.1 Pro is here — same core as Deep Think, built for everyday tasks. Now #1 on Artificial Analysis' Intelligence Index, beating Claude Opus at lower cost. ARC-AGI-2 at 77.1% (more than doubled from 31.1%). Google has quietly built the most capable practical AI model available today. If you use the Gemini app or access AI via API, this update is already working for you.