Google launches Gemini 3 with record-setting benchmarks

On November 18, 2025, Google unveiled its latest flagship large-language model, Gemini 3, and made it available at once across the Gemini app and AI-enhanced Search.

The new model family includes the 'Pro' variant for general use, and teases a 'Deep Think' version aimed at more advanced reasoning tasks.

What distinguishes the launch is the set of record-breaking benchmark scores that Gemini 3 Pro reportedly achieved — for example, on the LMArena leaderboard and other reasoning/science benchmarks it out-performed previous top models, including competitors such as GPT-5 and Claude Sonnet 4.5.

Google described the model as its 'most intelligent yet,' hinting at a step change in how AI can reason, plan, and interact across modalities (text, vision and tools).

Beyond the model itself, Google also introduced a new developer-oriented coding tool, 'Antigravity', built around Gemini 3’s agentic capabilities. This coding environment allows AI agents to access a code editor, terminal, and browser in tandem — effectively enabling multi-step coding workflows rather than single-prompt responses.

In practical terms, the rollout means that Gemini 3 Pro is immediately powering Google’s Search 'AI Mode' experience, the Gemini mobile app, the Gemini API through AI Studio/Vertex AI, and enterprise subscriptions — a level of integration that rivals the company’s earlier model launches.

For developers, Gemini 3 offers big leaps in tool use, long-horizon planning and multimodal reasoning. For example, Google claims major score gains in mathematics, coding benchmarks and spatial/multimodal reasoning tasks compared to prior models.

This launch arrives amid intense competition in the AI model race. Just days earlier, GPT‑5.1 from OpenAI and Grok 4.1 from xAI had been unveiled — Google’s rapid deployment of Gemini 3 underscores the company’s ambition to push AI beyond research labs into real-world product relevance.

In short, Gemini 3 is more than an incremental upgrade. It represents a strategic shift by Google: launching a top tier LLM at scale, integrating it across its product stack from day one, and introducing support for agentic workflows — tools that can not only generate text but also plan, act and interact with environments. Its success will likely be judged not only by benchmark-leadership, but by how widely it is adopted by developers, enterprises and end-users in the months ahead.