Google has commenced the deployment of its much-anticipated Gemini 3 model as of Tuesday, with the objective of outpacing competitors OpenAI and Anthropic. CEO Sundar Pichai characterized Gemini 3 as a “state-of-the-art” and “supremely intelligent” model, engineered to enhance capabilities in coding and navigating intricate queries.
The company has made Gemini 3 accessible for all users via the Gemini app for Google AI Pro and Ultra subscribers, within AI Mode in Search, for developers utilizing the Gemini API in AI Studio, Google Antigravity—its innovative agentic development platform, Gemini CLI, and for enterprises via Vertex AI and Gemini Enterprise.
Since its inception two years ago, Gemini has significantly integrated into the mainstream. The Gemini app boasts over 650 million users, while AI Overviews are utilized by approximately 2 billion people monthly.
Capabilities of Gemini 3
Google asserts that Gemini 3 is meticulously designed for robust multimodal comprehension and serves as a formidable model for “vibe coding.” The company claims that Gemini 3 surpasses its predecessor across all AI benchmarks, achieving a remarkable score of 1501 points on the LM Arena leaderboard, indicative of PhD-level reasoning, and receiving top accolades on Humanity’s Last Exam and GPQA Diamond.
The Gemini 3 Pro has also excelled in multimodal reasoning, attaining an impressive 81 percent on MMMU-Pro and 87.6 percent on Video-MMMU. This underscores the model’s versatility in addressing complex issues across a spectrum of disciplines, including the sciences and mathematics.
In a recent presentation, Google showcased Gemini 3’s ability to comprehend both visual and spatial data, producing pertinent outcomes. For instance, it can interpret and translate handwritten recipes into diverse languages, culminating in a shareable cookbook.
By feeding Gemini academic papers and video lectures, users can generate code to develop interactive flashcards, thereby facilitating a more engaging and visually appealing learning experience.
For developers, Google promotes Gemini 3 as the pinnacle of vibe coding and its most sophisticated agentic coding model, outstripping the former Gemini 2.5 Pro. Developers can leverage Gemini 3 within Google AI Studio, Vertex AI, Gemini CLI, and the newly minted Google Antigravity platform.
Google touts Gemini 3 as its most secure AI model, having undergone rigorous safety evaluations. “The model manifests reduced sycophancy, heightened resistance to prompt injection, and superior safeguards against cyberattack misuse,” the company asserted.
Revamped Interface: Dynamic View
Users will discern a modernization in the Gemini app, which has been meticulously revamped. Google has unveiled “generative interfaces,” an innovative interface type that emerges dynamically upon request. For instance, a query for “a three-day trip to Rome next summer” yields a visually stunning itinerary reminiscent of a magazine layout, complete with images and modules.
Additionally, a new experience layer called Dynamic View crafts and codes a unique user interface in real time. Asking Gemini to “elucidate the Van Gogh gallery with contextual insights for each artwork” will render an interactive response, inviting users to engage by tapping and scrolling through the content.
Newly integrated into the Gemini app is Gemini Agent, powered by Gemini 3. This tool efficiently manages multi-step tasks directly within the application, providing connectivity with Google apps, calendar management, reminder setting, prioritization of tasks, and even drafting responses. Currently, Gemini Agent is rolling out to Gemini AI Ultra users as a priority.
Google is also incorporating Gemini 3 into Search, commencing with AI Mode. By selecting “Thinking” from the model drop-down menu, Gemini 3 enables users to pose complex and nuanced inquiries, permitting the AI to better grasp context and intent prior to responding.
The rollout for Gemini 3 in AI Mode is now underway for all users in the US, with enhanced limits for those subscribed to Google AI Pro and Ultra plans.
Additionally, Google introduced a new agent platform entitled “Google Antigravity,” which allows developers to code at a higher, task-specific level.
Competitive Landscape in AI
This year has manifested a cutthroat competition in the quest for AI supremacy, with all major AI firms openly vying to construct superior and swifter large language models (LLMs).
Google is entering this arena, following the lead of OpenAI and Anthropic, who unveiled their own advanced models mere months prior. The efficacy of this confluence of speed, capability, and diverse features remains to be seen.
While Gemini 3 is hailed as a potent model equipped with excellent multimodal functionality and reasoning aptitude, Google is still under considerable pressure from both nascent and established players as the marketplace brims with frontier models increasingly adept at reasoning and tailored for autonomous tool application.
A pressing inquiry concerns who will ultimately seize dominion over the AI market and whether that leadership will manifest from the US, China, or another locale.
In January, Chinese newcomer DeepSeek astonished observers with its open-source model, suggesting a rising wave of formidable Chinese innovations ambitiously seeking global presence.
Concurrently, Mark Zuckerberg’s Meta has assembled a new superintelligence team, successfully attracting several top-tier OpenAI scientists, though the company’s strategic objectives remain elusive.
Source link: Indianexpress.com.
