Google’s Gemini 3 Flash goes global as default model, aiming straight at OpenAI

Google Gemini

Google has released Gemini 3 Flash, a new lightweight AI model designed to deliver high-end performance at lower cost and higher speed. Built on the Gemini 3 foundation announced last month, the Flash variant is now being set as the default model across the Gemini app globally, as well as AI Mode in Google Search.

The timing does seem to be deliberate. Gemini 3 Flash arrives just six months after Gemini 2.5 Flash, and the jump in performance is substantial. On several benchmarks, the new model closes the gap with frontier systems, including Google’s own Gemini 3 Pro and OpenAI’s GPT-5.2.

On Humanity’s Last Exam, a benchmark that tests expert-level reasoning across multiple domains, Gemini 3 Flash scored 33.7 percent without tool use. That puts it close to Gemini 3 Pro at 37.5 percent and slightly behind GPT-5.2 at 34.5 percent, while comfortably outperforming Gemini 2.5 Flash, which managed just 11 percent.

The gains are even more pronounced in multimodal reasoning. On the MMMU-Pro benchmark, which measures how well models understand and reason across text, images and other inputs, Gemini 3 Flash scored 81.2 percent, topping all competing models tested.

Google is betting that this blend of speed, cost efficiency and capability will make Gemini 3 Flash the default choice for everyday AI use. While users can still switch to Gemini 3 Pro from the model picker, particularly for demanding maths and coding tasks, Flash will now handle most queries by default.

The company is also pushing creative and practical use cases inside the Gemini app. Users can generate app prototypes using prompts, upload short videos for feedback, submit sketches for interpretation, or analyse audio recordings to extract insights or create quizzes. Google says the model is better at understanding user intent and can respond with richer visual outputs, including tables and images, rather than plain text answers.

On the enterprise side, Google claims early traction. Companies such as JetBrains, Figma, Cursor, Harvey and Latitude are already using Gemini 3 Flash through Vertex AI and Gemini Enterprise. Developers can access the model in preview via the API and through Antigravity, Google’s recently launched coding tool.

Gemini 3 Pro, meanwhile, is expanding its reach. It is now available to everyone in the US for Search, and Google is widening access to its Nano Banana Pro image model within Search as well. Google says Gemini 3 Pro scores 78 percent on the SWE-bench verified coding benchmark, second only to GPT-5.2, making it well suited for complex coding, video analysis, data extraction and visual question answering.

With Gemini 3 Flash, Google is making a clear statement. High-quality AI, it argues, does not need to be slow, expensive or locked behind premium tiers to compete at the top end.