For the past three months, Google’s Gemini 3 Pro has held its own as one of the most capable Frontier models available. But in the fast-moving world of AI, three months is a lifetime – and competitors aren’t standing still.
Earlier today, Google released Gemini 3.1 ProAn update that brings a significant innovation to the company’s workhorse power model: three levels of adjustable thinking that effectively turn it into a lightweight version of Google’s signature Deep Think reasoning system.
This release is the first time Google has released a "a point" The update to the Gemini models signals a change in the company’s release strategy from periodic full-version launches to more frequent incremental upgrades. More importantly for enterprise AI teams evaluating their model stack, 3.1 Pro’s new three-tier thinking system – low, medium, and high – gives developers and IT leaders a single model that can dynamically scale their reasoning effort, from quick responses for routine questions to multi-minute deep reasoning sessions for complex problems.
This model is now available in preview via the Gemini API Google AI StudioGemini CLI, Google’s agentic development platform AntiGravity, Vertex AI, Gemini Enterprise, Android Studio, consumer Gemini apps, and NotebookLM.
‘Deep Think Mini’ effect: adjustable logic on demand
The most consequential feature in Gemini 3.1 Pro isn’t a single benchmark number – it’s the introduction of a three-tier thinking level system that gives users granular control over how much computational effort the model invests in each response.
The Gemini 3 Pro only offered two thinking modes: low and high. The new 3.1 Pro adds a medium setting (same as the previous high) and, critically, what an overhaul it does "High" Meaning. When set to High, the 3.1 Pro behaves as a "Short version of Gemini Deep Think" – The company’s special logic model was Just updated last week.
The implications for enterprise deployments could be significant. Instead of routing requests to different specific models based on task complexity – a common but operationally cumbersome pattern – organizations can now use a single model endpoint and adjust logic depth based on the task at hand. Routine document summarization can run at low think with fast response times, while complex analytical tasks can be scaled up to high think for Deep Think-caliber reasoning.
Benchmark performance: More than double the Reason 3 Pro
Google’s published benchmarks tell a story of dramatic improvements, particularly in areas related to reasoning and agentic capabilities.
But ARC-AGI-2A benchmark that evaluates a model’s ability to solve novel abstract reasoning patterns, scored 3.1 Pro 77.1% – More than double the 31.1% achieved by the Gemini 3 Pro and well ahead of Anthropic’s Sonnet 4.6 (58.3%) and Opus 4.6 (68.8%). This result even beats OpenAI’s GPT-5.2 (52.9%).
The benefits are spread across the board. But final test of humanityA rigorous academic reasoning benchmark, 3.1 Pro achieved 44.4% without tools, higher than the 37.5% for 3 Pro and ahead of both Cloud Sonnet 4.6 (33.2%) and Opus 4.6 (40.0%). But GPQA DiamondOn a scientific knowledge assessment, 3.1 Pro reached 94.3%, outperforming all listed competitors.
Where the results become particularly relevant to enterprise AI teams is in agentic benchmarks – evaluations that measure how well models perform given tools and multi-step tasks, the type of work that increasingly defines production AI deployments.
But Terminal-Bench 2.0which evaluates agentive terminal coding, the 3.1 Pro scored 68.5% compared to its predecessor’s 56.9%. But mcp atlasA benchmark measuring multi-step workflows using the Model Context protocol, 3.1 Pro reached 69.2% – a 15-point improvement from 3 Pro’s 54.1% and about 10 points ahead of both Cloud and GPT-5.2. and on browsecompWhich tests agentive web search capability, the 3.1 Pro achieved 85.9%, surpassing the 3 Pro’s 59.2%.
Why Google chose a ‘0.1’ release – and what it signals
The versioning decision is noteworthy in itself. Previous Gemini releases had adopted a pattern of dated previews – for example, multiple 2.5 previews before reaching general availability. The choice to designate this update as 3.1 instead of the other 3 Pro Previews suggests that Google considers the improvements significant enough to warrant a version increase, while "a point" The framing sets the expectation that this is an evolution, not a revolution.
Google’s blog post states that 3.1 Pro is based directly on the lessons of the Gemini Deep Think series, incorporating techniques from both the earlier and more recent versions. The benchmarks strongly suggest that reinforcement learning played a central role in the gains, especially on tasks like ARC-AGI-2, coding benchmarks, and agentic evaluation – exactly the domains where RL-based training environments can provide clear reward signals.
The model is being released in preview rather than a general availability launch, with Google saying it will continue to make progress in areas like agentic workflows before moving to full GA.
Competitive Implications for Your Enterprise AI Stack
For IT decision makers evaluating frontier model providers, the release of Gemini 3.1 Pro will cause them to rethink not only which models to choose, but how to adapt to such a rapid pace of change for their own products and services.
The question now is whether this release generates a reaction from competitors. The original launch of the Gemini 3 Pro last November kicked off a wave of model releases in both proprietary and open-source ecosystems.
With 3.1 Pro regaining benchmark leadership in several key categories, the pressure is on Anthropic, OpenAI, and the Open-Weight community to respond – and in the current AI landscape, that response is measured in weeks, not months.
Availability
Gemini 3.1 Pro is now available in preview for developers via the Gemini API in Google AI Studio, Gemini CLI, Google AntiGravity, and Android Studio. Enterprise customers can access it through Vertex AI and Gemini Enterprise. Consumers on Google AI Pro and Ultra plans can access it through the Gemini app and NotebookLM.
<a href