On Tuesday, Google released Gemini 3, its latest and most advanced foundation model. The model is now immediately available through the Gemini app and AI search interface.
This release comes just seven months after the Gemini 2.5 release, positioning the new model as Google’s most capable LLM yet and an immediate contender for the most capable AI tool available. The launch follows closely after recent releases from competitors, including OpenAI’s GPT 5.1 last week and Anthropic’s Sonnet 4.5 just two months ago, highlighting the incredibly fast pace of development in frontier AI models.
A more research-intensive version of the model, called Gemini 3 Deepthink, will be made available to Google AI Ultra subscribers in the coming weeks. This version will be released once it completes further rounds of safety testing.
Google’s head of product for the Gemini model, Tulsee Doshi, commented on the new capabilities, stating that with Gemini 3 they are seeing a massive jump in reasoning. She noted that the model is responding with a level of depth and nuance that has not been seen before.
This reasoning power is already reflected in independent benchmarks. The model achieved a record high score of 37.4 on the Humanity’s Last Exam benchmark, which is designed to measure general reasoning and expertise. This surpasses the previous high score of 31.64, which was held by GPT-5 Pro. Gemini 3 also reached the top of the leaderboard on LMArena, a human-led benchmark that measures user satisfaction.
According to Google, the Gemini app currently has more than 650 million monthly active users. Additionally, 13 million software developers have used the model as part of their workflow.
Alongside the base model, Google also released a new Gemini-powered coding interface called Google Antigravity. This interface allows for multi-pane agentic coding, similar to agentic IDEs like Warp or Cursor. Specifically, Antigravity combines a ChatGPT-style prompt window with a command-line interface and a browser window that can show the impact of changes made by the coding agent.
DeepMind CTO Koray Kavukcuoglu explained that the agent can work with your editor, across your terminal, and across your browser to ensure it helps you build your application in the best way possible.