Google unveiled Gemini 3 Flash, a next-generation AI that is fast and affordable

After last year’s launch of Gemini 3 Pro, Google today officially announced Gemini 3 Flash — a new AI model for everyday users and developers.

The main message of Gemini 3 Flash — «advanced intelligence with a focus on speed and low cost». The model retains Gemini 3’s strengths, including sophisticated logical reasoning, multimodal and visual understanding, and high performance in agent-based and «vibe»-coding. At the same time, it offers «Flash-level» latency, energy efficiency and significantly lower cost. The Flash lineup remains the most popular among Google’s models.

Expectantly, the Gemini 3 Flash outperforms the Gemini 2.5 Flash across the board and «significantly» outperforms the Gemini 2.5 Pro in a number of tests. It’s comparable to the Gemini 3 Pro in terms of features, and even beats it in some benchmarks, including MMMU Pro, Toolathlon, and MCP Atlas.

Gemini 3 Flash outperforms the Gemini 2.5 Flash across the board, and ̲ significantly outperforms the Gemini 2.5 Pro in several tests.

• GPQA Diamond (Scientific Knowledge): 90.4%
• Humanity’s Last Exam (academic thinking): 33.7% without instruments
• MMMU Pro (multimodal understanding and reasoning): 81.2%
• SWE-Bench Verified (agent-based programming): 78%
• Toolathlon (long-term real-world programming challenges): 49.4%
• MCP Atlas (multi-step workflows with MCP): 57.4%

Google emphasizes that the Gemini 3 Flash «outperforms the 2.5 Pro while being three times faster and significantly cheaper». The model’s strengths —logic, tooling and multimodality — enable more complex tasks, including advanced video analysis, data extraction and visual Q&A. This is especially relevant for developers building support chatbots, gaming assistants and other AI agents.

The model can be used by developers building support chatbots, gaming assistants and other AI agents.

Gemini 3 Flash is already starting to roll out in the Gemini app and replaces 2.5 Flash as the default model. Google calls it «a major upgrade to everyday AI» combining «intelligence and speed». For example, users will be able to quickly create useful apps from scratch without programming knowledge — all they need to do is describe an idea and refine it with Gemini. It can also turn voice memos and thought flow into prototypes.

It’s notable that the Gemini 3 Flash will be available in two modes on the model list:

• Fast — for quick responses
• Thinking — for complex tasks

Gemini 3 Pro will still show up as «Pro» and focus on advanced math and programming tasks.

In addition, Gemini 3 Flash becomes the default model in AI Mode worldwide. It allows you to ask more nuanced questions and get detailed answers. In parallel, Google is making the Gemini 3 Pro with Generative UI and Nano Banana Pro available to all users in the U.S.

For developers, Gemini 3 Flash is already available in preview mode via AI Studio, Google Antigravity, Gemini CLI and Android Studio. The new product complements the Gemini 3 Pro and Gemini 3 Deep Think lineup unveiled earlier this month.