Google has revealed Gemini 3, a new generation of its AI model that puts the focus on the advanced reasoningContext understanding and multimodality. The company positions it as its most capable system to date, with broad deployment in consumer products, developer tools, and enterprise platforms.
Beyond the announcement, the real interest lies in how these improvements translate into everyday life: more concise answers, interactive elements in the search and programming capabilities and agents more robust. For Spain, there are two key points: Native support for Catalan, Basque, and Galician, and an availability plan that starts in different areas and will gradually reach our region.
What does Gemini 3 offer compared to previous versions?
Google's roadmap with Gemini has been cumulative: Gemini 1 expanded the quantity and variety of actionable information, Gemini 2 solidified the agency and reasoning, and 2.5 Pro climbed records in public rankings. With Gemini 3The firm integrates this advancement so that the system understand nuances and intention from the user's perspective, with fewer back-and-forths to reach the useful answer.
According to Sundar Pichai, the goal is to offer AI at "Google scale," combining infrastructure, research and product in a single stack: Gemini app, AI Studio, Vertex AI and new features such as the agent platform Google AntigravityThe company also highlights less "flattery" in the responses and a more direct style.
Availability and reach in Spain and Europe
Gemini 3 premieres from day one in the AI Search Modealthough that experience begins in United StatesIn Spain, it will be accessible through the Gemini app or APIs for developers through AI Studio and Vertex AI, with gradual deployment to other surfaces.
In the linguistic section, Google expands support with 30 new languagesincluding Catalan, Basque, and Galician. This expansion is key for equipment and users in Europe seeking working with local content, official documents and workflows in co-official languages.
Performance: public and internal tests
In synthetic metrics, Gemini 3 leads the ranking of LM Arena with 1.501 pointsIn academic reasoning, he scores 37,5% in Humanity's Last Exam (without tools) and 91,9% in GPQA Diamondand marks a remarkable 23,4% in MathArena Apex, results that aim to consolidate a more stable performance in complex problems.
Mode DeepThink This takes that calculation profile a step further: 41,0% in Humanity's Last Exam (without tools) and 45,1% in ARC-AGI 2 with code execution, a test that attempts to measure abstract reasoning ability and phased planningGoogle emphasizes that these are advances that, while based on benchmarks, aim to be translated into real-world tasks.
Multimodality and expanded context window
Gemini 3 is “multimodal by nature”: it processes text, image, audio and video combined, a point that the company considers differentiating. In internal evaluations, the model stands out with 81% in MMMU-Pro y 87,2% in Video-MMMU, and improves its factual accuracy to 72,1% in SimpleQA Verified.
The context window grows to 1 million tokensThis allows you to upload code repositories, extensive papers, or video lessons and work on that basisAmong the use cases cited: converting handwritten family recipes into a coherent book or transforming a scientific article into interactive cards and visualizations to study it better.
Programming and agents: Google Antigravity
In development, the numbers support: Gemini 3 leads the way WebDev Arena with 1.487 ELO, it obtains a 54,2% in Terminal-Bench 2.0 (usage of tools via terminal) and 76,2% in SWE-bench Verified, indicators that the model handles real-world environments and engineering tasks better.
The most striking novelty is Google AntigravityIt is an agent-based development platform with access to the editor, terminal, and browser. This enables the system plan and execute tasks end-to-end software and validate your own code, leaving the final review and audit to human teams.
In everyday use, Gemini 3 powers Gemini Agent in the Gemini app, which is capable of organizing Gmail mail, plan itineraries or run multi-step flows with browser control. Some advanced features are reserved for paid plans (AI Pro/AI Ultra) and will be rolled out gradually.
Search with interactive elements
With Gemini 3 integrated into AI Mode, the responses can include widgets, simulations and calculators in real time. If you search for “planetary orbits”, the system could generate one adjustable displayIf you need to compare loans, show a calculator parameterized with your data.
For those who don't want the heaviest model, the search engine offers a selection menu: an option faster by default for most uses, or the advanced model for queries that require deeper reasoning and interactive outputs.
Security and external verification
Google claims that Gemini 3 is their safest model To date, with less complacency in the responses, greater resistance to the prompt injections and stricter controls against misuse (for example, cyberattacks). These measures are part of its Frontier Safety Framework.
In addition to internal assessments, the company cites collaborations with independent bodies and experts, such as the AISI of the United Kingdomand third-party reviews (Apollo, Vaultis, Dreadnode). In Europe, this approach to testing and risk governance is key to regulated environments and deployments in companies.
Competitive context and roadmap
The launch comes amidst fierce competition from other leading models. Google maintains that Gemini 3 surpasses recent rivals in several classifications, especially in reasoning, multimodality, and programming, and prepares iterations such as DeepThink for even more demanding scenarios. Furthermore, this rivalry includes moves by platforms and manufacturers, such as the rapprochement between Apple and Google and the possible integration of Gemini into voice assistants, according to analyses on the new Siri.
The company notes that its ecosystem is maintaining traction: the AI-powered view reaches 2.000 million monthly usersThe Gemini app surpasses 650 millionOver 70% of cloud customers use AI capabilities, and there are already 13 million developers building upon their models.
With a rollout that begins in US Search and extends across apps and development platforms, Gemini 3 puts the emphasis on reasoning, agents, and personalizationThese pillars, if consolidated in real-world experience, can make a difference for users, technical teams, and companies in Spain and the rest of Europe.
