Google Unveils Gemini 3 Pro, Boosting AI Performance Across Products
Key Points
- Google launches Gemini 3 Pro, its most intelligent AI model yet.
- Gemini 3 Pro scores 37.5% accuracy on Humanity's Last Exam, beating the previous leader by 12.1 points.
- The model leads LMArena with a score of 1,501 points.
- New Gemini Agent feature lets the AI perform tasks like email management directly.
- Available in the Gemini app, Search AI Mode, and AI Overviews for AI Pro and Ultra subscribers.
- Developers can access the model through Gemini API, AI Studio, and Vertex AI.
- Antigravity coding app introduced for autonomous programming.
- Early safety testing of Gemini 3 Deep Think precedes rollout to AI Ultra users.
Google announced Gemini 3 Pro, its most intelligent AI model to date, offering stronger reasoning, coding, and multi‑modal understanding. The model set new benchmark scores, including a top accuracy of 37.5 percent on Humanity's Last Exam and a leading 1,501 points on LMArena. Gemini 3 Pro powers the Gemini app, AI Mode in Search, and new agentic features like Gemini Agent, enabling tasks such as email management. The model is also available through the Gemini API, AI Studio, and Vertex AI for developers and enterprise customers, with early access for AI Pro and Ultra subscribers.
New Model Capabilities
Google introduced Gemini 3 Pro as its most capable AI system, highlighting advances in reasoning, coding performance, and multi‑modal comprehension. The model delivers more concise, better‑formatted answers in the Gemini app and powers a new feature called Gemini Agent, which can execute user‑requested tasks directly, such as managing an email inbox.
Performance Benchmarks
Gemini 3 Pro achieved a top accuracy score of 37.5 percent on Humanity's Last Exam, surpassing the previous leader by 12.1 percentage points without relying on web‑search tools. On the LMArena benchmark, the model leads the leaderboard with a score of 1,501 points.
Integration Across Google Services
The model is rolling out in several Google products. In the Gemini app, users can select "Thinking" to access Gemini 3 Pro, with higher‑frequency limits for AI Plus, Pro, and Ultra subscribers. In Search, Gemini 3 Pro appears in AI Mode and AI Overviews, where it augments the fan‑out technique to retrieve more credible and relevant content, and can generate interactive interfaces like loan calculators. Gemini Agent requires permission to access Google apps to perform tasks.
Developer and Enterprise Access
Developers and enterprise customers can use Gemini 3 Pro via the Gemini API, AI Studio, and Vertex AI. A new agentic coding app called Antigravity, which autonomously programs and tracks its progress, is also being released. Early safety testing of an enhanced reasoning mode, Gemini 3 Deep Think, is underway before it becomes available to AI Ultra subscribers.