On Tuesday, Google launched Gemini 3, its new and most superior foundation model, that’s now instantly available via the Gemini app and AI search interface.
Coming simply 7-months after the Gemini 2.5 release, the brand new model is Google’s most capable LLM yet, and a instant contender for the most successful AI tool on the market. The launch also comes much less than a week after OpenAI launched GPT 5.1, and an insignificant 2-months after Anthropic launched Sonnet 4.5 — a reminder of the intense frontier model evolution.
A more research-rigorous version of the model, called Gemini 3 Deepthink, will also be made available to Google AI Ultra subscribers in the coming weeks, as soon as it passes in addition rounds of security testing.
“With Gemini 3, we’re observing this huge jump in reasoning,” stated Tulsee Doshi, Google’s head of product for the Gemini model. “It’s reacting with a level of depth and nuance that we haven’t seen before.”
Some of that reasoning power is already registering on independent benchmarks. With a rating of 37.4, the model marked the best score on record on the Humanity’s Last Exam benchmark, supposed to capture general reasoning and skill. The preceding high score, held by GPT-5 Pro, was 31.64. Gemini 3 also topped the leaderboard on LMArena, a human-led benchmark that measures user satisfaction.
As per Google, the Gemini app presently has more than 650 million monthly active users, and 13 million software developers have used the model as part of their workflow.
Beside the base model, Google also launched a Gemini-powered coding interface called Google Antigravity, permitting for multi-pane agentic coding much like agentic IDEs like Warp or Cursor 2.0. Particularly, Antigravity combines a ChatGPT-style activate window with a command-line interface and a browser window that can show the effect of the changes made by the coding agent.
“The agent can work with your editor, across your terminal, throughout your browser to make sure that it supports you construct that application within the high-quality way feasible,” stated DeepMind CTO Koray Kavukcuoglu.
