On Tuesday, Google officially launched Gemini 3, the newest and most powerful version of its foundation model series — one that the company says sets a new benchmark for reasoning, problem-solving, and real-time responsiveness. The model is now available immediately through the Gemini mobile app and through Google’s AI-powered Search interface, giving millions of users instant access to the upgraded system.
The launch comes just seven months after Gemini 2.5 and arrives during one of the most competitive periods in AI development. The rollout follows OpenAI’s recent announcement of GPT-5.1 and Anthropic’s release of Claude Sonnet 4.5, underscoring the breakneck pace at which frontier models are improving in 2025.
Google also confirmed that a more research-oriented version of the system, Gemini 3 Deepthink, will be launched in the coming weeks for subscribers of Google AI Ultra. That version is still undergoing intensive, multi-layered safety testing before it is made available to the public or enterprise developers.
“With Gemini 3, we’re seeing this massive jump in reasoning,” said Tulsee Doshi, head of product for the Gemini series at Google. “It’s responding with a level of depth and nuance that we haven’t seen before.”
Setting New Benchmark Records
Early performance results hint at a significant leap in capability. Gemini 3 achieved a record score of 37.4 on the highly demanding benchmark called Humanity’s Last Exam-HLE, which is designed to measure broad reasoning, domain-expert knowledge, and real-world decision competency. The previous record holder, GPT-5 Pro, scored 31.64.
It also ranked first in LMArena, a human-evaluated benchmark that measures user satisfaction and the perceived helpfulness of model outputs. Google says these results reflect sustained investment in long-context reasoning, safer outputs, and more natural conversational understanding.
The Growing Gemini Ecosystem
Google announced that the Gemini app now serves more than 650 million monthly active users, a milestone that reflects the rapid pace at which the model has been adopted since its original launch. More than 13 million developers have integrated Gemini into coding, automation, and creative workflows-a signal that the model is becoming deeply embedded in the worldwide software ecosystem.
In addition to the base model, Google also announced Antigravity, a new Gemini-powered coding environment geared toward competing more directly with next-gen agentic IDEs such as Cursor 2.0 and Warp. Antigravity combines a conversational AI prompt panel with a terminal interface and a live browser pane to create one unified space for a developer where an AI agent can write, execute, and test code all in one.
“The agent can work with your editor, across your terminal, across your browser,” explained DeepMind CTO Koray Kavukcuoglu, “ensuring it helps you build your application in the best way possible.”
What Gemini 3 Brings to Users
While Google has not yet released the complete set of architecture details, early testers and developers point out a number of key upgrades:
Improved multi-step reasoning for research, coding, and analytical tasks
Improved memory and context windows to handle large documents more effectively.
More natural, conversational style with less need for prompt engineering
Stronger safety guardrails and real-time toxicity filtering
Smoother integration across Google Workspace-including Docs, Sheets, and Gmail
If approved, Deepthink’s release is expected to further develop long-form reasoning and complex analysis, likely able to go head-to-head with OpenAI’s most advanced models in scientific, legal, and engineering circles.
A Rapidly Escalating AI Race
The launch of Gemini 3 puts a point on the high-stakes race among major AI labs to create increasingly sophisticated models that are capable of autonomy, advanced reasoning, and multi-modal understanding. With Google’s latest entry now live and Deepthink on the way, analysts expect even more aggressive competition in the next few months from OpenAI, Anthropic, Meta, and smaller startups building specialized reasoning engines.
For now, Google is signaling that Gemini 3 marks a turning point. “Every generation of these models gets us closer to AI that truly supports people in deeper, more meaningful ways,” said Doshi. “This is the biggest leap yet.”
