google launches gemini 3 with new coding app and record benchmark scores

Google Unveils Gemini 3: A New Era in AI Capabilities

Google officially launched Gemini 3 on Tuesday, representing its most sophisticated foundation model to date. Access to this advanced AI is now available directly through the Gemini application and integrated within the AI-powered search experience.

Rapid Advancement in AI Model Development

The release of Gemini 3 follows closely on the heels of Gemini 2.5, debuting just seven months prior. This new iteration is Google’s most powerful Large Language Model (LLM) yet, immediately positioning itself as a leading contender in the competitive AI landscape.

This launch occurs within a remarkably dynamic period, arriving less than a week after OpenAI’s release of GPT 5.1 and only two months following Anthropic’s introduction of Sonnet 4.5. These events underscore the incredibly rapid pace of innovation in frontier model development.

Gemini 3 Deepthink: Enhanced Reasoning Capabilities

A more extensively researched version of the model, known as Gemini 3 Deepthink, is slated for release to Google AI Ultra subscribers in the coming weeks. Its availability is contingent upon successful completion of further safety evaluations.

“Gemini 3 demonstrates a substantial leap in reasoning abilities,” stated Tulsee Doshi, Google’s Product Lead for the Gemini model. “The model exhibits a depth and subtlety in its responses that haven’t been previously observed.”

Benchmark Performance and User Engagement

Independent evaluations confirm this enhanced reasoning power. Gemini 3 achieved a score of 37.4 on the Humanity’s Last Exam benchmark, designed to assess general reasoning and expertise. This is a record-breaking result, surpassing the previous high score of 31.64 held by GPT-5 Pro.

Furthermore, Gemini 3 has secured the top position on the LMArena leaderboard, a benchmark that relies on direct user satisfaction ratings.

Currently, the Gemini app boasts over 650 million monthly active users. Additionally, more than 13 million software developers are actively incorporating the model into their development workflows.

Introducing Google Antigravity: A Coding Interface

Alongside the core model, Google introduced Google Antigravity, a Gemini-powered coding interface. This tool facilitates multi-pane, agentic coding, similar to IDEs like Warp or Cursor 2.0.

Antigravity integrates a ChatGPT-style prompt window with a command-line interface and a browser window. This allows developers to visualize the effects of code changes made by the AI agent in real-time.

“The agent is designed to seamlessly interact with your editor, terminal, and browser, ensuring it assists you in building applications as efficiently as possible,” explained DeepMind CTO Koray Kavukcuoglu.

This integrated approach aims to streamline the coding process and enhance developer productivity.