Google expanded its Gemini 3 family with Gemini 3 Flash, a model positioned as a faster and less expensive route to the reasoning and multimodal capabilities introduced with Gemini 3. The company said the release was rolling out globally across consumer, developer and enterprise products.
For consumers, Gemini 3 Flash became the default model in the Gemini app, replacing Gemini 2.5 Flash. Google also began deploying it as the default model for AI Mode in Search. The company presented those changes as a way to bring more capable reasoning to everyday tasks such as interpreting images, video and audio, creating quizzes and turning spoken ideas into simple applications. In Search, Google said the model could parse multifaceted questions, combine information with links and organize responses for complex tasks such as trip planning or learning unfamiliar subjects.
Developers received access through the Gemini API in Google AI Studio, Gemini CLI and Google's Antigravity development platform. The model was also made available through Android Studio, while enterprise customers could use it through Vertex AI and Gemini Enterprise. Google said Gemini 3 had been processing more than one trillion tokens a day through its API since the family launched.
Pricing for Gemini 3 Flash was set at $0.50 per million input tokens and $3 per million output tokens. Audio input remained priced at $1 per million tokens. Google said the model ran three times faster than Gemini 2.5 Pro in Artificial Analysis testing and used 30 percent fewer tokens on average than 2.5 Pro when operating at its highest thinking level on typical traffic. Those performance comparisons were supplied by Google and should be read as company claims tied to its selected tests and usage measurements.
Google also published benchmark results intended to show that the smaller, latency-focused model retained strong reasoning performance. It reported scores of 90.4 percent on GPQA Diamond, 33.7 percent on Humanity's Last Exam without tools and 81.2 percent on MMMU Pro. On SWE-bench Verified, a coding-agent benchmark, Google reported a score of 78 percent, above the results it listed for its Gemini 2.5 models and Gemini 3 Pro.
The release focused on workloads where response time matters alongside reasoning quality. Google highlighted iterative coding, tool use, video analysis, visual question answering and data extraction as likely applications. It also said JetBrains, Bridgewater Associates and Figma were already using the model. The announcement did not provide independent validation for all of its benchmark and customer-performance claims, but it established Gemini 3 Flash as Google's broad-access model across its main AI distribution channels.
