**Google Reveals Gemini 3: Advancing AI Functions**
Google has formally introduced Gemini 3, an innovative AI model crafted to help users “transform any concept into reality.” The initial version in this new lineup, Gemini 3 Pro, starts its rollout today, upgrading the Gemini app and AI Mode.
### Progression of Gemini
The Gemini series has experienced notable enhancements since its launch. Gemini 1.0 presented native multimodality and extended context windows. A year later, Gemini 2.0 introduced sophisticated reasoning and initial agentic functionalities, while Gemini 2.5 further refined reasoning and programming abilities. Now, Gemini 3, which omits the “.0” label, is heralded as Google’s “most clever model,” emphasizing context and intent comprehension with minimal cues.
### Main Attributes of Gemini 3 Pro
Gemini 3 Pro is designed to provide feedback that is “intelligent, succinct, and straightforward,” emphasizing authentic insights over platitudes. It serves as a thought collaborator, aiding users in grasping complex information and articulating ideas innovatively. Prominent features include translating elaborate scientific ideas into high-fidelity visuals and supporting creative brainstorming sessions.
### Performance Metrics
Gemini 3 Pro has achieved remarkable results across numerous benchmarks, exceeding its predecessor, Gemini 2.5 Pro, in all significant categories:
– **Humanity’s Last Exam:** 37.5% (without tools)
– **GPQA Diamond:** 91.9%
– **MathArena Apex:** 23.4% (new state-of-the-art)
– **MMMU-Pro:** 81%
– **Video-MMMU:** 87.6%
– **SimpleQA Verified:** 72.1%
– **WebDev Arena:** 1487 ELO
– **Terminal-Bench 2.0:** 54.2%
– **SWE-bench Verified:** 76.2%
These metrics demonstrate that Gemini 3 Pro excels in tackling complex challenges across various domains, including science and mathematics.
### Gemini 3 Deep Think Mode
In conjunction with Gemini 3 Pro, Google revealed the Gemini 3 Deep Think mode, which aims to provide even more enhanced reasoning and multimodal comprehension. This mode surpasses Gemini 3 Pro in essential evaluations, scoring 41.0% on Humanity’s Last Exam and 93.8% on GPQA Diamond. It will be accessible for AI Ultra subscribers in the upcoming weeks.
### Generative UI Features
Gemini 3 brings forth generative UI, enabling large language models (LLMs) to produce both content and entire user experiences. This breakthrough facilitates the automated creation of web pages, games, tools, and applications tailored to user prompts. The system functions with detailed guidelines and post-processing to improve output quality.
### Future Features and Integrations
Beginning this week, Gemini 3 will be incorporated into Google Search and AI Mode, allowing subscribers to opt for “Thinking: 3 Pro reasoning and generative layouts.” This integration enhances Google’s query fan-out technique, facilitating more nuanced searches for superior responses.
Moreover, Google intends to update its automatic model selection for subscribers, permitting them to direct challenging inquiries to Gemini 3 while employing quicker models for simpler tasks.
### Google Antigravity
Alongside Gemini 3, Google unveiled Antigravity, a new agentic development platform that enables developers to function at a higher, task-centric level. This IDE allows agents to independently plan and perform intricate software tasks across various settings.
### Conclusion
With the debut of Gemini 3, Google is establishing a new benchmark in AI capabilities, concentrating on multimodal reasoning, user experience personalization, and advanced problem resolution. As this technology progresses, it promises to enhance user interaction with AI, making it a formidable resource for creativity and productivity.
