Google announced the launch of Gemini 3, a new series of artificial intelligence models the company describes as its “most intelligent” and capable to date. The flagship model, Gemini 3 Pro, was made available immediately across several key Google products, including the Gemini app and Google Search. The launch signals a significant push by the company to move beyond text-based conversations and integrate AI more deeply into user workflows through interactive interfaces and automated, multi-step task execution.
In a blog post, Google and Alphabet CEO Sundar Pichai framed the release as the next chapter in the “Gemini era,” which began nearly two years prior. He highlighted the significant user adoption of Google’s AI features, stating that AI Overviews in Search now have 2 billion monthly users and the Gemini app surpasses 650 million monthly users. The release of Gemini 3 is positioned as a culmination of previous advancements, combining the multimodal understanding of Gemini 1 with the reasoning and agent-like capabilities introduced in Gemini 2.
Reasoning and multimodal understanding
According to Google, Gemini 3 Pro demonstrates state-of-the-art performance in reasoning, allowing it to grasp more depth and nuance in user requests. The company claims this results in more helpful and concise responses with less need for detailed prompting. A notable characteristic emphasized by Google is a change in the model’s tone; its responses are described as “smart, concise and direct, trading cliché and flattery for genuine insight.” The Verge noted this as a subtle jab at competitors like OpenAI’s ChatGPT, as Google also mentioned the model shows “reduced sycophancy,” an issue where models tend to agree with users excessively.
The model’s capabilities are “natively multimodal,” meaning it was designed from the ground up to process and synthesize information from various formats, including text, images, video, audio, and code, simultaneously. Google provided several examples of this in practice: a user could provide a video of their pickleball match for an analysis and a generated training plan, or upload photos of handwritten family recipes in different languages for the AI to decipher, translate, and compile into a shareable cookbook.
Multiple sources reported that Gemini 3 Pro achieved top scores on several independent AI benchmarks. VentureBeat and TechCrunch highlighted its leading position on the LMArena leaderboard, a platform that ranks models based on human preferences, where it was the first model to surpass a score of 1500 Elo. It also set new records on benchmarks measuring PhD-level reasoning and advanced mathematics. In terms of factual accuracy, a key concern for generative AI, Google stated that Gemini 3 Pro scored 72.1% on the SimpleQA Verified benchmark. The New York Times reported this figure, noting that while it is high for such a model, it still falls short of perfect accuracy.
Generative UI: AI that builds its own interface
A central innovation highlighted in the launch is the concept of “Generative UI,” or generative interfaces. Detailed in a paper from Google Research, this capability allows the AI model to generate not just content, but an entire interactive user experience tailored to a specific prompt. Instead of delivering information in a static block of text, Gemini 3 can design and code a custom visual layout on the fly.
This feature is being rolled out in two main products:
- In Google Search’s AI Mode, available to Google AI Pro and Ultra subscribers in the U.S., Gemini 3 can create dynamic experiences. For a query about a complex scientific topic like the “three-body problem,” it can generate an interactive simulation where users can manipulate variables. For a financial question about mortgages, it might build a custom loan calculator directly within the search results.
- In the Gemini app, this technology appears as two experiments called “visual layout” and “dynamic view.” Visual layout presents information in an immersive, magazine-style format, such as creating a visual travel itinerary for a trip to Rome. Dynamic view goes a step further by coding a unique, interactive interface for a prompt. An example provided was a request to see a gallery of Van Gogh’s paintings, which resulted in an interactive page where users could tap on artworks to learn more about them.
Google stated that these new interface types are designed to make information clearer, more actionable, and more engaging, allowing users to explore topics in greater depth.
Gemini Agent: A personal assistant for complex tasks
Building on the model’s improved reasoning and ability to use tools, Google introduced an experimental feature called Gemini Agent. This feature, initially rolling out to Google AI Ultra subscribers in the U.S. via the Gemini app, is designed to function as an autonomous agent that can orchestrate and complete complex, multi-step tasks on a user’s behalf.
By connecting to other Google apps like Gmail and Calendar, Gemini Agent can handle requests that require multiple actions. For instance, a user could ask it to “organize my inbox,” and the agent would prioritize important emails and draft replies for approval. A more complex prompt, such as booking a rental car for an upcoming trip, would involve the agent locating flight details from the user’s email, researching rental options that fit the budget and criteria, and preparing the booking for final confirmation. Google emphasized that the user remains in control, as the agent is designed to seek approval before taking critical actions like making a purchase.
Developer features and market context
While the primary focus was on consumer-facing features, Google also released tools for developers to build applications using the new model. A new platform named Google Antigravity was introduced to allow for a more collaborative, “agent-first” development experience. This is part of Google’s strategy to encourage a broader ecosystem of AI-powered applications.
The launch of Gemini 3 places Google in direct competition with other major players in the AI field, such as OpenAI and Anthropic, who have released their own model updates in recent months. The New York Times described the situation as an “A.I. arms race,” noting that the high cost of developing these powerful models is facing scrutiny from investors who question whether the business applications can justify the immense spending. However, Google’s ability to integrate Gemini 3 directly into products with massive existing user bases, like Search and the Gemini app, gives it a significant advantage in distribution. To further boost adoption, Google announced it would offer its Google AI Pro subscription service free for a year to college students in the U.S.
Sources: Google Blog, Google Blog, Google Blog, Google Research, The Verge, New York Times, TechCrunch, VentureBeat
