Gemini 3.0 Pro and Flash: Google’s answer to Sora 2

Following the release of Sora 2 from OpenAI, Google is lining up its answer: the official debut of the next-generation Gemini 3.0 is expected as soon as next week. Data from closed testing has surfaced online, and developers with early access to Gemini 3.0 Pro are already sharing striking real‑world examples.

Gemini 3.0 will arrive in Pro and Flash versions. In ARC-AGI-2 tests, the model posted accuracy above 20%, and on a so‑called “final exam” it reached a record 32.4%—higher than GPT‑5 and Grok 4. What stands out most is its programming aptitude: developers describe the Pro variant as a powerhouse for front‑end and web development. It assembles full interfaces, produces SVG graphics, and even builds physics simulations, from a ball tracing a complex path to simplified models of the solar system.

The Flash edition focuses on speed and precision. Insiders note its effectiveness on practical tasks, such as crafting travel itineraries. That said, even the Pro version isn’t flawless—generating accurate images of hands remains a stumbling block.

Despite these individual gaps, the leaks suggest Google is moving up a gear in the race for AI leadership. If the rumors bear out, the industry could be looking at one of the year’s most anticipated releases within days—an unveiling that feels less like catching up and more like a bid to set the pace.