Google announced during its Made by Google showcase that Gemini now has more than 1 billion monthly users. That is a big number by any measure, and Google says the app is the fastest-growing product in its history.
Voice is a large part of why people are coming back. Google says 63% of Gemini users speak to the assistant directly, and some rely on voice alone. The company also shared that busy parents are 43% more likely to use voice for everyday tasks. Image generation is also picking up, with Gemini producing more than 150 million images a day.
The app has passed 100 million active users on iOS. On macOS, people seem to be heavy users, sending prompts at about double the rate of other platforms. On Android, the assistant can now trigger actions inside more than 40 apps.
New models aim for cost and scale
Days before the user milestone, Google shipped three new models. Gemini 3.6 Flash is now the default workhorse, running about 17% leaner on output tokens and offering a context window above one million tokens. There is also Gemini 3.5 Flash-Lite and Gemini 3.5 Flash Cyber. Flash Cyber is the first Gemini model built to find and patch software vulnerabilities, but it is limited to governments and vetted partners for now.
Those launches help explain Google’s push for wider use, but independent testing tells a different story. On the Artificial Analysis Intelligence Index, Anthropic’s Claude Opus 5 leads with a score of 63. Google’s best-placed model, Gemini 3.6 Flash, scores 52. Several models from OpenAI, xAI, and Kimi also rank above or at the same level.
Google does have the fastest model on the Artificial Analysis board. Gemini 2.5 Flash-Lite has a time-to-first-response of 0.31 seconds. But on composite intelligence benchmarks that combine coding, reasoning, and knowledge, it falls behind. So while the app is growing quickly, the strongest model position still belongs to rivals.
More integrations and a sign language model
Google also said Gemini will connect to services such as Wix, OpenTable, Ticketmaster, Pandora, Angi, and Zocdoc over the coming weeks. That should make the assistant useful in more everyday settings.
Separately, Google DeepMind unveiled a sign-language-to-text model called SL2T. It was trained on more than 100,000 hours of data across over 50 sign languages. The model powers sign-to-text dictation in Gboard and Live Transcribe on the Pixel 11, starting with American Sign Language to English. The idea is to let Deaf users sign to their phone wherever they would normally type.
The user milestone is real, and the voice figures point to a shift in how people interact with assistants. But the benchmark gap shows growth and raw capability are not the same thing. Google is still chasing the next billion users, and it will need to close that gap if it wants to be seen as the leader in model quality too.
