(Credit: Google)
Days after OpenAI released GPT 5.1, Google is introducing its own, potentially more powerful AI model with Gemini 3, which starts rolling out to users today.
The company says Gemini 3 is Google’s most intelligent AI model. It hyped up Gemini 2.5 with the same language back in March, but this time, Google is confident enough to release Gemini 3 immediately to everyone. "This is the first time we are shipping Gemini in Search on day one,” says CEO Sundar Pichai.
Gemini 3 is rolling out via AI Mode, Google's ChatGPT-like interface on the Google search engine. One standout feature is how Gemini 3 can display “immersive visual layouts and interactive tools and simulations, all generated completely on the fly based on your query.”
The new model is also launching for all users in the Gemini app. And Google promises a boost for regular searches, too. For each query, the company plans to use Gemini 3 to perform “even more searches to uncover relevant web content” that may have been missed previously.
In addition, the company is signaling that Gemini 3 will eventually power AI Overviews, the Google Search function that automatically summarizes the answer to your query (for better or worse) at the top of the results page.
“In the coming weeks, we’re also enhancing our automatic model selection in Search with Gemini 3. This means Search will intelligently route your most challenging questions in AI Mode and AI Overviews to this frontier model—while continuing to use faster models for simpler tasks,” the company wrote in a separate blog post.
That said, the Gemini 3 integration with AI Overviews will initially be limited to paid subscribers of Google’s AI plans. In the US, they can also use a more powerful Gemini 3 Pro model starting today via AI Mode by selecting the “Thinking” option from the model drop-down menu.
According to Google, the Gemini 3 model outperforms GPT 5.1 and Anthropic’s Claude Sonnet 4.5 across a wide range of AI-related benchmarks, including math, scientific reasoning, and multilingual questions and answers.
“It’s state-of-the-art in reasoning, built to grasp depth and nuance—whether it’s perceiving the subtle clues in a creative idea, or peeling apart the overlapping layers of a difficult problem,” Pichai added.
(Credit: Google)Ironically, though, the same benchmarks also indicate Gemini 3 struggles with “visual reasoning puzzles” and “academic reasoning.” However, the Google model still beats the competition in these tests. The company also says Gemini 3 has been built to resist “sycophancy” and “prompt injection” attacks that can manipulate the AI into executing malicious instructions.
To attract software developers, the company also announced “Google Antigravity,” a suite of AI-powered tools for computer programming, an apparent response to rival coding programs from OpenAI and Anthropic.
“Using Gemini 3’s advanced reasoning, tool use, and agentic coding capabilities, Google Antigravity transforms AI assistance from a tool in a developer’s toolkit into an active partner,” Google’s CEO said. “Now, agents can autonomously plan and execute complex, end-to-end software tasks simultaneously on your behalf while validating their own code.”
In addition, Google has developed an even smarter “Gemini 3 Deep Think mode.” But the company wants to be careful with its release. “We’re taking extra time for safety evaluations and input from safety testers before making it available to Google AI Ultra subscribers in the coming weeks,” the company explained.