Google has introduced Gemini 1.0, a family of AI models built to combine text, images, audio, video and code. The 6 December 2023 announcement sets out three versions.
Ultra is the largest, aimed at demanding tasks. Pro is the general-purpose version, while Nano is designed to run on devices. Google says the models were trained across different types of information from the outset.
In its own evaluations, Google reports that Ultra exceeds previous leading results on 30 of 32 widely used academic benchmarks. Those results describe Ultra, rather than the Pro version entering Bard.
What you can use now
A specially tuned Gemini Pro is rolling out in Bard in English across more than 170 countries and territories. The initial experience accepts text prompts; support for other types of input is still planned. Google says Europe and additional languages will follow.
Ultra is not part of that public rollout. Google plans to introduce it through Bard Advanced in early 2024, after further safety checks and a tester programme. The model family’s multimodal design therefore goes beyond what Bard users can try at launch.
A smaller model for phones
On Pixel 8 Pro, Nano powers summaries in the Recorder app, including without a network connection. Summaries are initially English-only. Smart Reply in Gboard is also starting as a developer preview, using the US English keyboard and initially supporting WhatsApp.
Google’s Android announcement describes AICore, an Android 14 system service that manages the model, runtime and safety features. Developers interested in building with Nano can apply for early access. Additional devices are planned, rather than included in this first release.
Developer access follows next week
Google schedules Gemini Pro API access for 13 December through Google AI Studio and Vertex AI. That is a separate step from the Bard upgrade: the developer API is not part of the 6 December release.
Discover more from Sulat AI & Tech News
Subscribe to get the latest posts sent to your email.
