Google Rolls out Gemini 3 with new coding app and record-breaking benchmark scores

Written by: Mane Sachin

Published on:

Follow Us

Google’s Release of Gemini 3 and Its Advancements

Google introduced Gemini 3 on Tuesday, unveiling its most advanced foundation model to date. The model is now instantly accessible through the Gemini app and Google’s AI-driven search interface.

Arriving only seven months after Gemini 2.5, this new version is described as Google’s strongest large language model so far, positioning it as a major competitor in the rapidly evolving AI landscape. Its debut follows closely behind OpenAI’s release of GPT-5.1 and Anthropic’s rollout of Sonnet 4.5, underscoring how quickly top-tier models are being developed.

Google also announced a more research-focused variant called Gemini 3 Deepthink, which will launch for Google AI Ultra subscribers in the coming weeks after it undergoes additional safety evaluations.

“We’re seeing a significant leap in reasoning with Gemini 3,” said Tulsee Doshi, Google’s head of product for the Gemini lineup. “It delivers depth and subtlety in its responses that we haven’t seen before.”

Early benchmark results reflect that improvement. With a score of 37.4, Gemini 3 achieved the highest result ever on the Humanity’s Last Exam reasoning benchmark, surpassing the previous top score of 31.64 held by GPT-5 Pro. It also ranked first on LMArena, a human-evaluated benchmark measuring user satisfaction.

User Adoption and Antigravity Coding Interface

Google shared that the Gemini app has surpassed 650 million monthly active users, and more than 13 million developers have integrated Gemini models into their development workflows.

Alongside the core model release, Google introduced a coding environment called Google Antigravity. The interface supports multi-pane, agent-driven development similar to systems like Warp or Cursor 2.0. It blends a conversational prompt panel, a command line, and a browser window so users can see real-time results from the coding agent.

DeepMind CTO Koray Kavukcuoglu explained that the agent works seamlessly across the editor, terminal, and browser, enabling it to help developers build applications more efficiently and effectively.

Also Read:

Stack Overflow is transforming itself into an AI data provider

Intuit Signs $100M+ Deal with OpenAI to Bring Apps to ChatGPT

Mane Sachin

My name is Sachin Mane, and I’m the founder and writer of AI Hub Blog. I’m passionate about exploring the latest AI news, trends, and innovations in Artificial Intelligence, Machine Learning, Robotics, and digital technology. Through AI Hub Blog, I aim to provide readers with valuable insights on the most recent AI tools, advancements, and developments.

For Feedback - aihubblog@gmail.com