Google Unveils Gemini 3 Flash: Free, Lightning‑Fast, and Outperforms Its Predecessor

Google announced the unexpected launch of Gemini 3 Flash, a large language model that combines the intelligence of Gemini 3 Pro with the speed of the Flash series and a dramatically lower price.

Speed and Low Cost

The Flash series is traditionally associated with high speed but modest reasoning; Gemini 3 Flash breaks that stereotype. Google states the model reaches "Frontier Intelligence" while pushing cost and latency to the limit. It is three times faster than the previous Gemini 2.5 Pro and costs only $0.5 per million input tokens and $3 per million output tokens, far cheaper than GPT‑4o or Claude 3.5 Sonnet.

Performance Beats Previous Pro

Despite the low price, Gemini 3 Flash outperforms its predecessor on several hard benchmarks.

Scientific Knowledge (GPQA Diamond): 90.4% accuracy, exceeding 2.5 Flash (82.8%) and 2.5 Pro (86.4%).

Code Ability (SWE‑bench Verified): 78.0% score, beating 2.5 Pro and slightly surpassing Gemini 3 Pro (76.2%).

Multimodal (MMMU‑Pro): 81.2% score, showing a large improvement in image and video understanding.

For example, when a complex app‑operation video is fed to the model, it can recognize UI changes, identify elements, and infer the underlying interaction logic, a capability that is especially valuable for front‑end development and testing.

The model is also adept at writing code, interpreting complex charts, and answering deep scientific questions.

Who Can Use It?

The model is available immediately and free of charge. Regular users see Gemini 3 Flash as the default model in the Gemini app or web interface. Developers can access it through Google AI Studio or the Gemini API.