Google today launched Gemini 3.1 Pro, its most advanced artificial intelligence model designed to solve highly complex, multi-step tasks. The new model shows a significant leap in core reasoning capabilities, achieving more than double the performance of its predecessor, Gemini 3 Pro, on a key industry benchmark. This upgraded AI is now rolling out to consumers, developers, and businesses across various Google platforms, marking a focused upgrade to its reasoning engine.[mlq+8]
Advanced Reasoning for Tough Problems
Gemini 3.1 Pro is built to handle problems where simple answers are not enough, offering enhanced intelligence for science, research, and engineering challenges. Google says the model is a smarter and more capable foundation for intricate problem-solving, building on the breakthroughs made possible by Gemini 3 Deep Think. It features improved software engineering behavior and overall usability, with better agentic capabilities in areas like finance and spreadsheet applications. Agentic capabilities allow the AI to perform complex, multi-step tasks with minimal manual intervention, acting more like an intelligent assistant.[case+8]
The model achieved a verified score of 77.1% on the ARC-AGI-2 benchmark, a difficult industry test that measures an AI's ability to solve entirely new logic patterns. This score highlights its substantial progress in understanding and applying abstract reasoning, moving beyond memorized patterns to tackle novel challenges. In competitive tests, Gemini 3.1 Pro outperformed several other leading models, including Claude Opus 4.6, Claude Sonnet 4.6, and GPT-5.2, across 12 different benchmarks. It also scored 44.4% on Humanity's Last Exam, a challenging reasoning test, surpassing Claude Opus 4.6 at 40.0% and GPT-5.2 at 34.5%.[mlq+12]
Multimodal Understanding and Creative Applications
A core strength of Gemini 3.1 Pro is its native multimodal understanding, meaning it can process and comprehend information from many different sources simultaneously. This includes diverse inputs such as text, audio, images, video, PDFs, and even entire code repositories. The model can handle vast datasets and challenging problems from these varied information streams, allowing for a more comprehensive analysis and response. For example, it can process up to approximately 8.4 hours of audio, or up to 1 million tokens, for tasks like summarization, transcription, and translation.[deepmind+6]
This enhanced intelligence translates into powerful practical applications across many fields. For instance, Gemini 3.1 Pro can generate website-ready, animated Scalable Vector Graphics (SVGs) directly from simple text prompts. Because these animations are built in pure code rather than pixels, they remain crisp and clear at any scale and maintain incredibly small file sizes compared to traditional videos. An example showed it generating a detailed SVG of a pelican riding a bicycle, complete with realistic body structure and bicycle details.[mlq+9]
The model also excels at interactive design. It can write code to generate complex 3D simulations, such as a starling flock flight simulation. It can create immersive experiences where users can manipulate the flock with hand-tracking technology and listen to a generative soundtrack that changes based on the birds' movements. This capability provides a powerful way for researchers and designers to rapidly prototype sensory-rich interfaces.[case+5]
Google DeepMind's official X account featured how a Google UX engineer utilized Gemini 3.1 Pro to develop a realistic city planning application. The model demonstrated the ability to handle complex terrains, draw infrastructure maps, simulate traffic flows, and finally generate high-quality visual effects for the city layout. Another impressive application involved Gemini 3.1 Pro building a real-time aerospace dashboard, successfully configuring a public telemetry data stream to visualize the orbital trajectory of the International Space Station. It also showcased creative coding by translating literary themes into functional code. One notable demonstration even depicted the model generating a lightweight Windows 11 WebOS interface with complete application icons, a well-structured start menu, and basic window interaction logic, making it resemble a runnable operating system.[eu+7]
Broader Access and Developer Tools
Gemini 3.1 Pro is now accessible to a wide range of users globally, starting February 20, 2026. Consumers with Google AI Pro and Ultra subscriptions can use the model with higher usage limits in the Gemini app and NotebookLM. Free Gemini users can also ask a limited number of questions to the model.[mlq+7]
For developers, the model is available in preview through the Gemini API in Google AI Studio, Gemini CLI, the agentic development platform Google Antigravity, and Android Studio. Businesses and enterprises can access Gemini 3.1 Pro in preview via Vertex AI and Gemini Enterprise. Google is releasing the model in preview to validate these updates and continue making further advancements in areas such as ambitious agentic workflows before its general availability.[mlq+12]
The model also introduces an expanded "thinking_level" parameter, now including a "MEDIUM" option. This allows developers to optimize trade-offs between cost, performance, and speed based on the complexity of the task. For instance, a "Low" setting could be used for simple tasks like autocomplete, "Medium" for more involved processes like code review, and "High" for complex debugging or problem-solving scenarios. The massive 1 million token context window is particularly significant for developers and researchers, allowing the model to process extremely long documents, entire codebases, or extended multimedia inputs, providing a deep contextual understanding for complex tasks.[docs+6]
Competitive Pricing and Safety Focus
Google has priced the preview version of Gemini 3.1 Pro competitively, maintaining the same tiered billing model as its predecessor, Gemini 3 Pro. For prompts up to 200,000 tokens, the input price is $2 per million tokens, and the output price is $12 per million tokens. For larger prompts exceeding 200,000 tokens, the input price increases to $4 per million tokens, and output to $18 per million tokens. This pricing structure makes Gemini 3.1 Pro less than half the price of a competitor like Claude Opus 4.6, despite achieving very similar benchmark scores in many areas.[digitalapplied+11]
Google emphasized its commitment to responsible AI development and safety with this release. Gemini 3.1 Pro underwent extensive security testing under Google's Frontier Safety Framework. Evaluations covered five key risk areas: chemical, biological, radiological, and nuclear (CBRN) information risks; cybersecurity vulnerabilities; risks of harmful manipulation; machine learning research ethics; and misalignment issues where AI behavior deviates from human intent. The model remained below critical thresholds in all these areas, indicating a strong focus on safety. It also showed slight improvements in text-to-text and multilingual safety compared to Gemini 3.0 Pro. As a preview product, Gemini 3.1 Pro is available "as is" and may have limited support, as outlined in Google's Pre-GA Offerings Terms.[trendingtopics+4]
The release of Gemini 3.1 Pro underscores Google's rapid pace in AI innovation, building on feedback since the launch of Gemini 3 Pro in November. The company continues to push towards more sophisticated agentic workflows that can perform complex multi-step tasks with minimal manual input. This new model is poised to empower users and developers to tackle their most challenging problems with greater intelligence, efficiency, and a deeper understanding of complex information.[blog+6]
