News
ReleasesOVHcloud unveils Olala mini, its first in-house LLMChinaA 9B model beats a 1T model on FinQA, trained with Crusoe Managed RLReleasesThe internal metaphysics of an LLMReleasesMapping Language Model Internal Representations to Physical Laws and Scientific ReasoningResearchCan China keep rogue AI in check? Meet the start-up testing the top modelsFrontierMicrosoft aims to challenge OpenAI, Anthropic with cheaper, specialized AI modelsOpen modelsMicrosoft rolls out new first-party models and updates to agentic AI continuous improvementFrontierMicrosoft unveils AI models for coding, cybersecurity as it takes on Anthropic, OpenAIFrontierGoogle is making it easier to opt out of Gemini AI model training – here’s howFrontierOpenAI Unveils New Astra AI Model After Scrapping Prior ReleaseReleasesOVHcloud unveils Olala mini, its first in-house LLMChinaA 9B model beats a 1T model on FinQA, trained with Crusoe Managed RLReleasesThe internal metaphysics of an LLMReleasesMapping Language Model Internal Representations to Physical Laws and Scientific ReasoningResearchCan China keep rogue AI in check? Meet the start-up testing the top modelsFrontierMicrosoft aims to challenge OpenAI, Anthropic with cheaper, specialized AI modelsOpen modelsMicrosoft rolls out new first-party models and updates to agentic AI continuous improvementFrontierMicrosoft unveils AI models for coding, cybersecurity as it takes on Anthropic, OpenAIFrontierGoogle is making it easier to opt out of Gemini AI model training – here’s howFrontierOpenAI Unveils New Astra AI Model After Scrapping Prior Release
EveryLLM
imgwavvidGemini 3 Pro

Gemini 3 Pro

Google DeepMind's Gemini 3 Pro, a multimodal model with a 1 million-token context window.

Gemini 3 Pro is a multimodal AI model from Google DeepMind, released in preview in 2025 as the third generation of the Gemini line. The company says it outperforms Gemini 2.5 Pro on major benchmarks in reasoning, multimodality and coding, and it ships across Google products including Search, the Gemini app, AI Studio and Vertex AI.

Share on WhatsApp

  • 1 million-token context window across text, images, video, audio and code
  • Google reports 1501 Elo on the LMArena leaderboard
  • 37.5% on Humanity's Last Exam without tools; 91.9% on GPQA Diamond
  • Deep Think reaches 45.1% on ARC-AGI-2 with code execution
  • First Gemini generation shipped in AI Mode in Search on day one

What it is

Gemini 3 Pro is the first model of the third generation of Google's Gemini family, put out in preview by Google DeepMind. It is a general-purpose multimodal model that reads and writes text, images, video, audio and code, with a context window of 1 million tokens. Google positions it for learning, building and planning, and highlights reasoning, tool use, agentic work and what it calls vibe coding. The company reports that it beats Gemini 2.5 Pro across major AI benchmarks, and it is offered through the Gemini app, AI Studio, Vertex AI, Google Antigravity and AI Mode in Search.

How it came about

Google says the Gemini era began nearly two years before this release, with Gemini 1 bringing native multimodality and long context, and Gemini 2 laying groundwork for agentic capabilities and reasoning. Gemini 2.5 Pro topped the LMArena leaderboard for over six months. Gemini 3 was announced by Google and Alphabet CEO Sundar Pichai together with Google DeepMind CEO Demis Hassabis and CTO Koray Kavukcuoglu. Gemini 3 Pro launched in preview across Google products on day one, and a Deep Think mode was given to safety testers ahead of release to Google AI Ultra subscribers.

Inside the model

Google describes Gemini 3 as built to grasp depth and nuance and to infer the intent behind a request with less prompting. It handles text, images, video, audio and code together, and combines reasoning, vision and spatial understanding with multilingual performance and a 1 million-token context window. The company says it is stronger at tool use and agentic tasks, and that its replies are concise and direct rather than flattering. No parameter count, layer layout, training data or compute figures are published here.

What it can do

Google reports these scores for Gemini 3 Pro: 1501 Elo on the LMArena leaderboard, 37.5% on Humanity's Last Exam without tools, 91.9% on GPQA Diamond, 23.4% on MathArena Apex, 81% on MMMU-Pro, 87.6% on Video-MMMU and 72.1% on SimpleQA Verified. Deep Think is reported at 41.0% on Humanity's Last Exam without tools, 93.8% on GPQA Diamond and 45.1% on ARC-AGI-2 with code execution. Google gives examples such as translating handwritten recipes, generating visualisation code and analysing sports footage.

How to use it

Gemini 3 Pro is available in preview through the Gemini app, AI Studio, Vertex AI, Google Antigravity and AI Mode in Search. Deep Think mode was first shared with safety testers and is planned for Google AI Ultra subscribers. The weights are proprietary and not downloadable.

Safety and testing

Google says Gemini 3 Deep Think was given to safety testers before wider release. No further evaluation or red-teaming detail is published here.

Reception and debate

Google's own announcement presents the model as its most intelligent yet and reports leading benchmark results.

Why it matters

Google frames Gemini 3 as the next step after Gemini 1's multimodality and long context and Gemini 2's agentic foundations, and as part of its push towards artificial general intelligence. The company reports 2 billion monthly users for AI Overviews, 650 million monthly users for the Gemini app and 13 million developers building with its generative models. Gemini 3 is the first Gemini generation shipped in Search on day one.

Questions people ask about Gemini 3 Pro

What is Gemini 3 Pro?

It is the first model of Google DeepMind's third Gemini generation, released in preview in 2025. It is a general-purpose multimodal model aimed at reasoning, coding and agentic work, and it is offered through Google's own apps and developer platforms rather than as downloadable weights.

How much can it take in at once?

Google gives Gemini 3 a context window of 1 million tokens. That lets it work through long material such as academic papers, video lectures and tutorials, and combine text, images, video, audio and code in a single task.

How does it differ from Gemini 2.5 Pro?

Google says Gemini 3 Pro outperforms 2.5 Pro on every major AI benchmark it reports, with gains in reasoning, multimodality and coding. Gemini 2.5 Pro had topped the LMArena leaderboard for more than six months before this generation.

What is Deep Think?

Deep Think is an enhanced reasoning mode for Gemini 3. Google reports higher scores than Gemini 3 Pro on Humanity's Last Exam, GPQA Diamond and ARC-AGI-2, and it went to safety testers first, with access planned for Google AI Ultra subscribers.

Where can I use it?

Google shipped Gemini 3 Pro in the Gemini app, AI Studio, Vertex AI, its Antigravity agentic development platform and AI Mode in Search. It was the first Gemini generation to reach Search on the day of its announcement.

Good reading and sources

    Where this page comes from

    There is no English Wikipedia article on Gemini 3 Pro yet, so the facts on this page come from the sources listed above: blog.google/products/gemini/gemini-3/, deepmind.google/models/model-cards/gemini-3-pro/. Some are the subject's own pages, so its claims about itself are reported as its claims. EveryLLM wrote the text in its own words, with AI assistance, and compares it with the sources to make sure no passage is copied.

    It may contain mistakes, and figures change over time. Nothing here is investment or buying advice. If something is wrong, tell us. Last refreshed 2026-11-14.

    The weekly digest

    Every Monday: the model of the day, new pages on EveryLLM, and the conferences and anniversaries of the week ahead.