Google launches Gemini 3: 7 things you need to know

O Google launched the Gemini 3, marking an aggressive leap in the race of artificial intelligence. The new model is not an incremental upgrade — it is a rewrite of AI’s reasoning capabilities, with performance that outperforms the company’s own predecessor and major competitors in complex scientific and mathematical benchmarks.

For the first time, a Gemini model debuts simultaneously on Google Search, the Gemini app, and the developer API, signaling the company’s push for immediate scale.

We have gathered 7 important points about Gemini 3

1. He understands what you want — without you explaining it often

Gemini 3 is designed to capture the context and intent behind what you ask. This means less rewriting of prompts, less frustration trying to find the right words. You describe an idea in a vague way, and he already understands the nuances of what you need.​

In effect, the template automatically suggests relevant follow-up questions — ones you’re likely to ask next. It’s like talking to someone who is actually paying attention.

2. Available now — but there is no date for arrival in Brazil yet

Gemini 3 Pro is now available in the United States through the Gemini app, Google Search (AI Mode), and the API for developers at Google AI Studio and Vertex AI. Google confirmed to G1 that the model “begins to be made available from today to all users”, but has not yet detailed the international expansion schedule.​

For the Brazilian public, the expectation is that the Gemini 3 will follow the same gradual release pattern as previous models, which historically arrive a few weeks after the American launch. The Gemini application has been available in Brazil since 2024, with support for Portuguese, and the country is part of the list of more than 150 territories covered by global expansion

As for paid plans, Brazil already has access to AI Pro (R$ 97/month) and AI Ultra (R$ 1,209.90/month, with a 50% discount in the first three months). Both plans can be shared with up to five people, and Brazilian students get 15 free months of AI Pro. The expectation is that Gemini 3 will be integrated into these plans as soon as it arrives in the country.

3. Create custom interfaces for each question

One of Gemini 3’s most visible innovations is “Dynamic View” — the ability to generate complete, interactive interfaces tailored to each response. When you ask for a travel plan, it doesn’t deliver a text list: it generates an interactive itinerary with maps, schedules and direct links to reservations.​

If you ask a child about quantum physics, the interface will be colorful and visual. The same question for an adult generates scientific graphs, equations and links to papers. In Google Search, this translates to interactive mortgage calculators or physical simulations that you manipulate in real time.

4. Performs complete tasks — not just answers questions

Gemini 3 brings “long-horizon planning,” allowing the model to execute multi-step workflows. This means they can book local services by browsing websites, filling out forms and confirming appointments — like scheduling a haircut or booking a car service.​

It can also organize your Gmail inbox: categorize emails, create draft replies, archive old threads, and highlight what really requires attention. Each critical action undergoes your approval, keeping you in control of the process

5. Deep Think comes to problems that require deep reasoning

The Gemini 3 Deep Think is a variation of the flagship model designed for tasks that require prolonged thinking. It doesn’t respond instantly — it processes longer, exploring multiple strategies before delivering the solution.​

The results are impressive: 41% on Humanity’s Last Exam (a test created to challenge even experts with a PhD) and 45.1% on ARC-AGI-2, a benchmark that measures the ability to solve completely new problems, without prior training. Deep Think will be made available in the coming weeks exclusively to AI Ultra plan subscribers.

6. Deep Think Mode for Time-Intensive Problems

The Gemini 3 Deep Think is a variation of the flagship model designed for tasks that require prolonged thinking. It doesn’t respond instantly — it processes longer, exploring multiple thinking strategies in parallel before delivering the answer.

The results are remarkable: 41% on Humanity’s Last Exam (compared to 37.5% on Pro), 93.8% on GPQA Diamond and, most impressively, 45.1% on ARC-AGI-2, a test that measures the ability to solve new problems without prior training. This last number is significant because ARC-AGI was developed precisely to detect “artificial general intelligence” — the ability to adapt to completely new challenges

Deep Think will arrive in the coming weeks, exclusively for AI Ultra plan subscribers.

7. What’s Next — Flash and Global Expansion

Google has confirmed that Gemini 3 Flash, a speed-optimized version, is on the way. While Pro focuses on advanced reasoning and long tasks, Flash will be designed for real-time applications, responding instantly with a lower computational cost.

Additionally, the company promises to expand Gemini 3’s geographic access beyond the United States in 2025, along with enhanced multimodal capabilities — including real-time video processing and integration with geospatial data.​

The timing of the launch is no accident. With OpenAI readying GPT-5 and Anthropic moving forward with Claude, Google needs to prove that its AI infrastructure — combining Model, Search, and productivity tools — offers a more integrated ecosystem than the competition. Gemini 3 is the company’s bet that advanced reasoning, autonomous agents and global-scale deployment can win this race.