Google Introduces Gemini 3 Pro

Google has opened preview access to Gemini 3 Pro, the company’s new flagship language model. The announcement was made in an official blog post.

According to the system card, the latest version outperforms leading competitors in mathematics, text analysis and coding tasks. Gemini 3 Pro Preview shows particularly strong results in software development benchmarks.

The model was trained from scratch using Google’s custom AI accelerators. Benchmark data shows Gemini 3 Pro leading across nearly every test except SWE-Bench Verified, where it trails Claude Sonnet 4.5 by a single percentage point. In several other benchmarks, Google’s model pulls ahead by a wide margin.

What’s New in Gemini 3 Pro

Google says the new release sets a higher bar for AI performance, thanks to advanced reasoning techniques. Responses are now more helpful, better formatted and more concise.

Gemini 3 Pro also delivers the strongest multimodal understanding Google has shipped so far. It excels at analyzing uploaded homework photos or decoding messy lecture notes. Coding inside Canvas has become more capable as well, showcasing notable improvements in “vibe coding” workflows.

The model can now generate journal-style output that blends images and text. A new “dynamic view” feature allows Gemini to produce interactive, scrollable responses. For example, turning a prompt about a Van Gogh gallery into a mini-website with selectable tabs.

Another major addition is Gemini Agent, which can directly interact with Google apps to take actions on the user’s behalf. It can triage your inbox, surface booking options for rentals based on your email, and more.

Google says Gemini 3 Pro will roll out to the Gemini app and Google Workspace (Docs, Sheets and Slides) within the next month. Paid subscribers will receive expanded usage limits.