Google has reignited the AGI competition with the launch of Gemini 3, a major upgrade the company says outperforms OpenAI’s GPT-5.1 and Anthropic’s Claude Sonnet 4.5 across several key benchmarks. The tech giant is touting advances on LMArena, Humanity’s Last Exam, GPQA Diamond and other reasoning tests — signaling one of Google’s strongest pushes in the frontier model race yet. Sundar Pichai described the system as “state-of-the-art in reasoning,” underscoring the company’s renewed momentum.
Alongside the model, Google introduced Google Antigravity, a new agent-centric development platform built tightly around Gemini 3. The tool gives AI agents access to an integrated editor, browser and terminal, enabling them to independently plan and execute software tasks. The approach resembles a fusion of Cursor, Replit and autonomous agent frameworks — but fused directly into Google’s ecosystem for deeper control and scalability.
DeepMind leaders Demis Hassabis and Koray Kavukcuoglu said this marks “the beginning of the Gemini 3 era,” noting that the model is already powering Google Search, Workspace, the Gemini app and the company’s development tools.
Gemini 3 brings improvements in multimodality, expanded context windows, enhanced reasoning modes and stronger safety performance. The messaging from Google makes it clear: the company sees this as another step on its path toward AGI.
Whether Gemini 3 can surpass GPT-5.1 in real-world workflows remains the deciding factor. But one thing is certain — Google has shifted the competitive landscape, and it wants the industry to take notice.
