What’s new in Google’s Gemini 3? A breakdown of the next-gen AI

Google has introduced Gemini 3, its newest and most advanced artificial intelligence model, as the company accelerates efforts to build AI systems that can think, plan and respond more like human assistants.

The next-generation model arrives nearly two years after Google introduced the first iteration of Gemini, built in response to the rapid rise of OpenAI’s ChatGPT in late 2022.

Google and Alphabet CEO Sundar Pichai said Gemini 3 combines the company’s prior advances in multimodality, reasoning and agentic capabilities and will be available “at the scale of Google”.

Google said the newest features powered by Gemini 3 will first be available to Gemini Pro and Ultra subscribers in the United States before a wider global rollout.

What is Gemini 3?

Gemini 3 is the third generation of Google’s flagship AI models, designed to understand and respond to information across text, images, audio and video. It builds on Gemini 1 and 2, adding deeper reasoning, stronger multimodal understanding (spanning text, images, audio and video), and new “agentic” capabilities that allow AI systems to operate more independently.

Google says it is its “most intelligent” model so far, capable of understanding context and intent with fewer prompts and providing more accurate, concise answers.

The launch comes nearly two years after Google began what it calls the “Gemini era,” aimed at competing directly with advanced systems from OpenAI and other AI labs.

What makes Gemini 3 different?

Google says Gemini 3 is a major step forward in several areas.

The model outperforms previous versions on a wide range of academic and industry benchmarks, including tests designed to measure logic, problem-solving, and accuracy.

Gemini 3 can interpret and analyse complex scientific diagrams, videos and mixed-format information more effectively than earlier models, according to the company.

Google said Gemini 3 is designed to be “smart, concise and direct”, offering insight rather than flattery. “It acts as a true thought partner,” they wrote.

Where will people see it?

Google Search: AI Mode will use Gemini 3 to generate more visual and interactive responses.

Gemini app on phones: The updated model will power conversations and daily tasks.

Gemini 3 will be available to developers via Google AI Studio, Vertex AI, the Gemini API and a new agentic development platform called Google Antigravity, which Google said lets AI agents act more autonomously inside an integrated development environment

Google added that the higher-performance “Deep Think” mode will be released to safety testers before being offered to Google AI Ultra subscribers in the coming weeks.

What is Gemini 3 Deep Think?

Deep Think is a high-performance mode for tougher reasoning tasks. According to Google, Deep Think performs even better than Gemini 3 Pro on advanced logic benchmarks and can solve more novel, open-ended challenges.

What does it mean for developers?

Google says Gemini 3 is its strongest model yet for coding. It can operate more independently inside development environments, writing, debugging and validating code with less supervision.

The new Antigravity platform allows the AI to use tools like a web browser or terminal on its own.

Why does this matter for Google?

Gemini 3 represents Google’s most aggressive push yet to keep pace with OpenAI, Meta and Anthropic in the rapidly accelerating AI industry.

The release of Gemini 3 comes months after OpenAI launched its fifth-generation ChatGPT model and Anthropic upgraded its Claude AI system, both of which have increasingly taken on the role of “answer engines” capable of tackling complex queries.

The model’s benchmarks and early integrations suggest Google is focused on closing the gap on reasoning, strengthening multimodal capabilities, boosting developer adoption, and expanding agentic workflows, which many in the industry view as the next big leap beyond chatbots.