How Will Google Gemini 3 Deliver on Agentic AI Promise?

Google has released Gemini 3, marking the third generation of its multimodal AI model since the company began the Gemini era two years ago. 

The model has launched across Google Search, the Gemini app, and developer platforms including AI Studio, Vertex AI, and the newly announced Google Antigravity development platform.

“It's amazing to think that in just two years, AI has evolved from simply reading text and images to reading the room,” says Sundar Pichai, CEO of Google and Alphabet. “Starting today, we’re shipping Gemini at the scale of Google.”

The Gemini app now serves 650 million monthly users, whilst AI Overviews reaches 2 billion users each month. Google says that 13 million developers have built applications using its generative models, with 70% of Cloud customers using the company's artificial intelligence products.

Google Gemini 3 advances reasoning and context awareness

Sundar said Gemini 3 combines reasoning capabilities with improved context awareness. “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,” he said. “Gemini 3 is also much better at figuring out the context and intent behind your request, so you get what you need with less prompting.”

Google has integrated Gemini 3 into AI Mode within Search, enabling what the company describes as generative user interface experiences. These include interactive tools, simulations and visual layouts generated in response to user queries.

Demis Hassabis, CEO of Google DeepMind, and Koray Kavukcuoglu, CTO of Google DeepMind and Chief AI Architect at Google, described the release as advancing the path towards artificial general intelligence. “It’s the best model in the world for multimodal understanding and our most powerful agentic and vibe coding model yet, delivering richer visualizations and deeper interactivity – all built on a foundation of state-of-the-art reasoning,” they said.

Gemini 3 Pro performance exceeds predecessor on benchmarks

Gemini 3 Pro achieves a score of 1,501 Elo on the LMArena Leaderboard, outperforming its predecessor, Gemini 2.5 Pro, which held the top position for over six months. The model scores 37.5% on Humanity’s Last Exam without tool usage and 91.9% on GPQA Diamond, tests that measure reasoning at PhD level. On MathArena Apex, the model reaches 23.4%, establishing a new benchmark for mathematics performance amongst frontier models.

For multimodal reasoning, Gemini 3 Pro scores 81% on MMMU-Pro and 87.6% on Video-MMMU. The model achieves 72.1% on SimpleQA Verified, a benchmark measuring factual accuracy.

Building on these capabilities, Google is introducing Gemini 3 Deep Think, an enhanced reasoning mode designed to extend the base model's performance. In testing, Gemini 3 Deep Think scores 41.0% on Humanity's Last Exam without tools and 93.8% on GPQA Diamond. The system achieves 45.1% on ARC-AGI-2 with code execution, demonstrating performance on tasks the model has not encountered during training. Gemini 3 Deep Think will undergo additional safety evaluations before becoming available to Google AI Ultra subscribers in the coming weeks.

Gemini Agent brings task automation to consumers

For consumer applications, Google has introduced Gemini Agent, available to Google AI Ultra subscribers. The feature connects to Google Calendar, Gmail and Reminders, executing tasks such as inbox organisation and schedule management. The system breaks tasks into steps, displays progress, and requires user approval before proceeding with actions.

The model maintains a 1 million-token context window, enabling processing of documents, video lectures and tutorials. Applications include generating interactive study materials, analysing handwritten recipes across languages, and creating training plans based on video analysis of athletic performance. These capabilities extend across both consumer and enterprise use cases.

“Every generation of Gemini has built on the last, enabling you to do more,” said Sundar. “Gemini 1’s breakthroughs in native multimodality and long context window expanded the kinds of information that could be processed – and how much of it. Gemini 2 laid the foundation for agentic capabilities and pushed the frontiers on reasoning and thinking, helping with more complex tasks and ideas, leading to Gemini 2.5 Pro topping LMArena for over six months.”

Demis and Koray said the company plans to release additional models in the Gemini 3 series. “We plan to release additional models to the Gemini 3 series soon so you can do more with AI,” they said. “We look forward to getting your feedback and seeing what you learn, build and plan with Gemini.”