Google launched Gemini 3 Flash with fast speed and low cost

Gemini 3 Flash is developed based on the Gemini 3 model released last month. The new version is deployed by default in the Gemini app globally and in the search engine’s AI Mode, replacing its predecessor Gemini 2.5 Flash.

Google said the new model is suitable for video analysis, data extraction, visual Q&A, and fast and repeatable workflows thanks to its high speed.

For software developers, the price of Gemini 3 Flash is 0.50 USD for one million input tokens (data units) and 3 USD for one million output tokens. Compared to Gemini 2.5 Pro, the new model performs better, runs three times faster, and uses 30% fewer tokens in thinking tasks. Gemini 3 Flash also scored 78% in SWE-bench Verified, a test suite that measures an AI system’s ability to write software, better than Gemini 3 Pro (76.2%), Gemini 2.5 Flash (60.4%), and only behind GPT-5.2 (80%).

“We position Flash as a powerful, reliable model. If you consider the cost of entry and exit, Flash is clearly a much cheaper option, allowing many companies to handle high-volume tasks,” Tulsee Doshi, Senior Director of Gemini models at Google, explained to Google. TechCrunch.

For regular users, Google says, Gemini 3 Flash is good at recognizing multimodal content and responding accordingly. For example, people can upload short videos playing pickleball and asking for advice, sketches for the model to guess the object in the drawing, or audio recordings to be analyzed.

Gemini 3 Flash also better understands user intent in queries and provides visual feedback with elements such as images and tables. Users can even use the new model to design application prototypes right in the Gemini application via command prompts.

Google released Gemini 3 Flash just one day after OpenAI launched GPT Image 1.5, an image generator with better instruction compliance, more precise editing, and image creation speeds four times faster than the previous version. Gemini 3 Flash marks Google’s latest effort to challenge OpenAI in the context of the increasingly fierce artificial intelligence race.