Gemini’s Strategic Shift: Why Accounts are Closing in the UK, EU, and Australia

Lisa Ernst · 14.02.2026 · Artificial Intelligence · 12 min

The artificial intelligence landscape is in constant flux, with new developments emerging at an astonishing pace. As an observer of this dynamic field, I find the advancements in models like Gemini particularly compelling. These innovations are not just theoretical; they are actively reshaping industries, influencing how we work, conduct research, and interact with information. While the focus of this article is on the broader impact of Google’s Gemini AI, it is important to address a recent, albeit separate, development concerning the cryptocurrency exchange also named Gemini.

Source: Own illustration

The cryptocurrency exchange Gemini, founded by Cameron and Tyler Winklevoss, has announced a strategic realignment. This involves a new round of layoffs and a decision to withdraw from the UK, EU, and Australian markets to concentrate on its operations in the United States. This move, which will see accounts in these regions closed by April 6, 2026, is part of a broader strategy to streamline operations, reduce costs, and accelerate profitability. The company cited advancements in artificial intelligence as a key factor in its workforce reduction, noting that AI has significantly boosted productivity across both technical and non-technical roles, allowing for efficient operations with fewer employees. Gemini also highlighted its continued investment in prediction markets, including its Gemini Predictions platform, which launched in mid-December and has already attracted over 10,000 users trading more than $24 million. Following this announcement, Gemini’s stock experienced a drop of over 7%, settling at around $7.

Now, let's pivot to the advancements in Google DeepMind's Gemini AI, which is driving innovation across various sectors. Google DeepMind recently unveiled Gemini 2.0, an AI model designed for what it terms the "agentic era," emphasizing its ability to understand and process real-world interactions. This development signifies a shift towards more sophisticated AI agents capable of performing multi-step tasks and acting on behalf of users.

Quick Summary

• Gemini Crypto Exchange Exits UK, EU, Australia: The cryptocurrency exchange Gemini is closing accounts in these regions by April 6, 2026, to focus on the US market, streamline operations, and reduce costs, partly due to increased AI-driven productivity.
• Google DeepMind's Gemini AI Advancements: Google DeepMind continues to develop its Gemini AI models, including Gemini 2.0, 2.5, and 3, focusing on "agentic" capabilities, multimodal understanding, and advanced reasoning.
• New Models and Features: Gemini 2.0 Flash (now deprecated) and newer versions like Gemini 3 Flash offer enhanced speed, multimodal inputs/outputs, and native tool integration.
• Agentic AI Projects: Initiatives like Project Astra (universal AI assistant) and Project Mariner (human-agent interaction) explore the future of AI agents.
• Enterprise and Government Adoption: Gemini is being integrated into Google Workspace, "Gemini for Government," and various enterprise solutions for automation, research, and customer service.
• Scientific Breakthroughs: Gemini Deep Think is achieving significant results in mathematics, physics, and computer science research, even reaching IMO gold medal standards.
• Developer Ecosystem: The Gemini CLI, new extensions, and open protocols (A2A, ACP) are expanding opportunities for developers and businesses.

Key Advancements in Gemini Models

Gemini 2.0 represents a significant leap forward, offering enhanced capabilities with native image and audio output, along with robust tool usage. It builds upon the foundational progress made by earlier versions, Gemini 1.0 and 1.5, which excelled in multimodal understanding and processing information across text, video, images, audio, and code. The model’s advanced reasoning capabilities are now being integrated into AI overviews, allowing for the handling of more complex topics and multi-step questions. These advancements are a direct result of decades of investment in Google’s comprehensive, full-stack approach to AI innovation.

One of the latest iterations, Gemini 2.0 Flash, was available experimentally for all Gemini users and developers through the Gemini API in Google AI Studio and Vertex AI. This model was designed for speed and scalability, outperforming its predecessor, 1.5 Pro, in key benchmarks with twice the speed. It supported multimodal inputs across images, video, and audio, and could generate native images with text and controllable text-to-speech in multiple languages. Furthermore, it could natively invoke tools like Google Search, execute code, and integrate custom third-party functions. Google also introduced a new Multimodal Live API, enabling real-time audio and video streaming inputs and the simultaneous use of multiple tools. However, it is important to note that Gemini 2.0 Flash and 2.0 Flash-Lite are scheduled for deprecation and will be shut down on March 31, 2026, as newer, more capable models emerge.

The Agentic Future: Project Astra and Mariner

Google extensively explores "agentic experiences" with Gemini 2.0, including initiatives like Project Astra and Project Mariner. Project Astra is a research prototype exploring the capabilities of a universal AI assistant, aiming for seamless integration into Google products. This assistant can converse in multiple and mixed languages, demonstrating improved understanding of accents and uncommon words. With Gemini 2.0, Project Astra leverages Google Search, Lens, and Maps, and has enhanced its ability to remember information for up to 10 minutes within a session.

Source: gigazine.net

Project Astra, an early prototype, uses Gemini 2.0 to power a universal AI assistant, aiming to integrate seamlessly into everyday Google products with enhanced language understanding.

Project Mariner, another early research prototype developed with Gemini 2.0, focuses on advancing human-agent interaction by processing information directly from browser screens. This allows it to perform tasks via an experimental Chrome extension. Both projects prioritize responsible AI development, with Google committing to rigorous safety and security protocols. This includes extensive "red teaming" exercises to test the models' reasoning capabilities and efforts to ensure agents prioritize user instructions over potential prompt injections.

Gemini in the Enterprise and Government

The influence of Gemini extends beyond research and developer tools, making significant inroads into enterprise and governmental sectors. Google supports the U.S. government's modernization efforts by introducing "Gemini for Government". This comprehensive offering integrates Google’s commercial cloud solutions, Gemini models, and agentic solutions, providing components such as enterprise-grade Google Search, video and image generation capabilities, and AI agents for deep research and idea generation. Priced at less than $0.50 per government agency annually, this initiative features FedRAMP High-authorized security and compliance, aligning with the GSA's OneGov strategy.

Source: vecteezy.com

Google supports the U.S. government with "Gemini for Government," integrating cloud solutions and AI agents for research and generation, all with FedRAMP High-authorized security.

In the commercial landscape, Gemini Enterprise acts as a central hub for workplace AI, combining advanced Gemini models with a no-code workbench, pre-built Google agents, and secure connections to enterprise data. This framework is supported by a robust governance structure and a partner ecosystem. Leading companies across various industries, including Banco BV, Behr, Box, DBS Bank, Deloitte, Deutsche Telekom, Fairprice Group, and the U.S. Department of Energy, leverage Google AI products. For example, Banco BV automated analytics for its customer service representatives using Gemini Enterprise, while Harvey, an AI tailored for legal and professional services, runs on Gemini.

Multimodal agents powered by Gemini are directly integrating into Workspace applications, understanding and generating text, images, video, and speech. Google Vids, for instance, transforms presentations into engaging videos with AI-generated scripts and voiceovers, serving 2.5 million users monthly. Google Meet now offers real-time voice translation for business clients, and a new Data Science Agent in preview automates data cleaning and ingestion, a tool already adopted by clients like Morrisons and Vodafone. The Customer Engagement Suite, a conversational AI solution, drives chatbots like Commerzbank’s Bene, which handles over two million chats and resolves 70% of inquiries. Mercari, Japan’s largest online marketplace, expects a 500% ROI by using Google AI to reduce customer service workload by at least 20%.

The Gemini for Developers Ecosystem

For developers, Google continues to expand the Gemini ecosystem. The Gemini CLI, an AI agent for terminal-based interaction, has seen over a million developers use it within three months of its launch. New Gemini CLI extensions allow for customization and connection to services. Open standards like the Agent2Agent Protocol (A2A) facilitate communication between agents, and the Agent Commerce Protocol (ACP) ensures secure financial transactions by agents.

Numerous companies are integrating Gemini into their products and services. Klarna uses Gemini and Veo for customized lookbooks, increasing orders by 50%. Mercedes-Benz incorporates Google AI into its MBUX Virtual Assistant, enabling cars to converse with drivers. Swarovski employs Vertex AI for personalized customer experiences, boosting email open rates by 17%. The Gemini model family generated over 13 billion images and 230 million videos, with companies like Figma using the Gemini Flash 2.5 Image model for high-quality, on-brand image creation. Virgin Voyages utilizes Veo's "text-to-video" and Imagen to produce hyper-personalized ads and emails. Google also partners with industry leaders such as Box, OpenText, Salesforce, ServiceNow, and Workday, along with consulting firms like BCG, Capgemini, and McKinsey, to help clients plan, deploy, and develop agents.

Gemini Deep Think: Advancing Scientific Research

Gemini Deep Think represents a vanguard in AI-driven scientific research, capable of solving complex problems in mathematics, physics, and computer science. By summer 2025, an advanced version of Gemini Deep Think achieved gold medal standards in the International Mathematical Olympiad (IMO) and similar results in the International Collegiate Programming Contest. This AI has since evolved to tackle more complex, open-ended challenges across various scientific and engineering disciplines.

Collaborative publications underscore Deep Think’s multidisciplinary impact. For pure mathematics research, a mathematical research agent, codenamed Aletheia, driven by Gemini Deep Think, uses a natural language verifier to identify errors and iteratively generates and refines solutions. This agent can admit failures, enhancing efficiency for human researchers, and navigates complex research via Google Search and web browsing to prevent incorrect citations and computational inaccuracies. Deep Think’s rapid progress since July 2025 is evident from its up to 90% score on the IMO-ProofBench Advanced test, with scaling laws extending to PhD-level exercises. Aletheia has autonomously solved several open mathematical problems, demonstrating its capacity for autonomous research, contributing to generalization in studies like BKKKZ26, and providing intermediate suggestions for works such as FYZ26 and ACGKMP26.

Source: dreamstime.com

Gemini Deep Think’s mathematical research agent, Aletheia, has achieved gold medal standards in the International Mathematical Olympiad, autonomously solving complex problems and collaborating with human researchers.

In computer science and physics, Gemini Deep Think has identified effective "recipes" for collaboration, particularly the "Advisor" model, using tactical techniques like "balanced prompting" and code-assisted verification. An advanced version supported the review of CS theory papers for the STOC’26 conference, helping solve long-standing bottlenecks in algorithms, ML, and combinatorial optimization. It broke deadlocks in classical computer science problems like "Max-Cut" and "Steiner Tree" by employing advanced continuous mathematics tools. Additionally, it constructed a combinatorial counterexample that disproved a decade-old conjecture in online submodular optimization, analyzed equations to prove a new ML optimization technique’s success through adaptive penalty generation, and expanded an economics theory theorem for AI using advanced topology and order theory. In physics, Gemini found a novel solution for calculating gravitational radiation from cosmic strings using Gegenbauer polynomials. These findings demonstrate how AI fundamentally transforms research by acting as a powerful scientific companion, augmenting human intellect through knowledge retrieval and rigorous verification.

The Gemini Model Lineup: Current and Future Versions

Google DeepMind offers a range of Gemini models, each optimized for different applications. The Gemini 3 models, representing Google DeepMind’s most intelligent AI models, include Gemini 3 Flash and Gemini 3 Pro. Gemini 3 Flash aims to accelerate idea realization, while Gemini 3 Pro stands as the world’s best model for multimodal understanding, excelling in agentic and "vibe coding" (a blend of creativity and coding in the AI context). Both support text, image, video, audio, and PDF inputs, with text output. Notably, Gemini 3 Pro supports code execution, function calling, search, and structured outputs, with token limits up to 1,048,576 for input and 65,536 for output. An Image Preview version of Gemini 3 Pro also supports image generation.

For cost-efficiency and high-volume processing, Gemini 2.5 Flash and Gemini 2.5 Flash-Lite offer versatile capabilities. Gemini 2.5 Flash is ideal for large-scale processing and low-latency, high-volume tasks requiring reasoning, while Flash-Lite is optimized for speed and throughput. Gemini 2.5 Pro remains the advanced reasoning model, capable of solving complex problems in code, mathematics, and STEM fields, and can analyze large datasets, codebases, and documents with a long context window. It supports audio, images, video, text, and PDF as inputs. It is important to note that Gemini 2.0 Flash and 2.0 Flash-Lite are set to be deprecated and shut down on March 31, 2026, as newer, more capable models emerge.

Source: the-decoder.com

Google DeepMind offers a range of Gemini models, each optimized for different applications, from the efficient Gemini 2.5 Flash to the advanced reasoning capabilities of Gemini 2.5 Pro.

Gemini Model Comparison Table

Conclusion

The Gemini suite of models, from the advanced reasoning of Gemini Deep Think to the versatile capabilities of Gemini 3 Flash, highlights a profound shift in the application of AI. These tools are empowering developers, transforming enterprise operations, aiding government initiatives, and fundamentally reshaping scientific research. As Google continues its commitment to responsible AI development, the "agentic era" promises further integration of intelligent agents into our daily lives and professional workflows, driving unprecedented innovation across a multitude of sectors.