For the past three months, Google’s Gemini 3 Pro has held its ground as one of the most capable frontier models available. But in the fast-moving world of AI, three months is a lifetime — and competitors have not been standing still.
Earlier today, Google released Gemini 3.1 Pro, an update that brings a key innovation to the company’s workhorse power model: three levels of adjustable thinking that effectively turn it into a lightweight version of Google’s specialized Deep Think reasoning system.
The release marks the first time Google has issued a “point one” update to a Gemini model, signaling a shift 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 scale its reasoning effort dynamically, from quick responses for routine queries up to multi-minute deep reasoning sessions for complex problems.
The model is rolling out now in preview across the Gemini API via Google AI Studio, Gemini CLI, Google’s agentic development platform Antigravity, Vertex AI, Gemini Enterprise, Android Studio, the consumer Gemini app, and NotebookLM.
The ‘Deep Think Mini’ effect: adjustable reasoning on demand
The most consequential feature in Gemini 3.1 Pro is not a single benchmark number — it is the introduction of a three-tier thinking level system that gives users fine-grained control over how much computational effort the model invests in each response.
Gemini 3 Pro offered only two thinking modes: low and high. The new 3.1 Pro adds a medium setting (similar to the previous high) and, critically, overhauls what “high” means. When set to high, 3.1 Pro behaves as a “mini version of Gemini Deep Think” — the company’s specialized reasoning model that was updated just last week.
The implication for enterprise deployment could be significant. Rather than routing requests to different specialized models based on task complexity — a common but operationally burdensome pattern — organizations can now use a single model endpoint and adjust reasoning depth based on the task at hand. Routine document summarization can run on low thinking with fast response times, while complex analytical tasks can be elevated to high thinking for Deep Think–caliber reasoning.
Benchmark Performance: More Than Doubling Reasoning Over 3 Pro
Google’s published benchmarks tell a story of dramatic improvement, particularly in areas associated with reasoning and agentic capability.
On ARC-AGI-2, a benchmark that evaluates a model’s ability to solve novel abstract reasoning patterns, 3.1 Pro scored 77.1% — more than double the 31.1% achieved by Gemini 3 Pro and substantially ahead of Anthropic’s Sonnet 4.6 (58.3%) and Opus 4.6 (68.8%). This result also eclipses OpenAI’s GPT-5.2 (52.9%).
The gains extend across the board. On Humanity’s Last Exam, a rigorous academic reasoning benchmark, 3.1 Pro achieved 44.4% without tools, up from 37.5% for 3 Pro and ahead of both Claude Sonnet 4.6 (33.2%) and Opus 4.6 (40.0%). On GPQA Diamond, a scientific knowledge evaluation, 3.1 Pro reached 94.3%, outperforming all listed competitors.
Where the results become particularly relevant for enterprise AI teams is in the agentic benchmarks — the evaluations that measure how well models perform when given tools and multi-step tasks, the kind of work that increasingly defines production AI deployments.
On Terminal-Bench 2.0, which evaluates agentic terminal coding, 3.1 Pro scored 68.5% compared to 56.9% for its predecessor. On MCP Atlas, a benchmark measuring multi-step workflows using the Model Context Protocol, 3.1 Pro reached 69.2% — a 15-point improvement over 3 Pro’s 54.1% and nearly 10 points ahead of both Claude and GPT-5.2. And on BrowseComp, which tests agentic web search capability, 3.1 Pro achieved 85.9%, surging past 3 Pro’s 59.2%.
Why Google chose a ‘0.1’ release — and what it signals
The versioning decision is itself noteworthy. Previous Gemini releases followed a pattern of dated previews — multiple 2.5 previews, for instance, before reaching general availability. The choice to designate this update as 3.1 rather than another 3 Pro preview suggests Google views the improvements as substantial enough to warrant a version increment, while the “point one” framing sets expectations that this is an evolution, not a revolution.
Google’s blog post states that 3.1 Pro builds directly on lessons from the Gemini Deep Think series, incorporating techniques from both earlier and more recent versions. The benchmarks strongly suggest that reinforcement learning has played a central role in the gains, particularly on tasks like ARC-AGI-2, coding benchmarks, and agentic evaluations — exactly the domains where RL-based training environments can provide clear reward signals.
The model is being released in preview rather than as a general availability launch, with Google stating it will continue making advancements in areas such as agentic workflows before moving to full GA.
Competitive implications for your enterprise AI stack
For IT decision makers evaluating frontier model providers, Gemini 3.1 Pro’s release has to not only make them rethink which models to choose but also how to adapt to such a fast pace of change for their own products and services.
The question now is whether this release triggers a response from competitors. Gemini 3 Pro’s original launch last November set off a wave of model releases across both proprietary and open-weight ecosystems.
With 3.1 Pro reclaiming benchmark leadership in several critical categories, the pressure is on Anthropic, OpenAI, and the open-weight community to respond — and in the current AI landscape, that response is likely measured in weeks, not months.
Availability
Gemini 3.1 Pro is available now in preview through the Gemini API in Google AI Studio, Gemini CLI, Google Antigravity, and Android Studio for developers. 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.
Credit: Source link
