Gemini, Google’s recently unveiled GenAI model family, is now available for Google Cloud customers on Vertex AI after its recent introduction to Bard and the Pixel 8 Pro. The Gemini Pro, a lighter version of the more powerful Gemini Ultra, currently in private preview for select customers, is now in public preview on Vertex AI. This is facilitated through the new Gemini Pro API, which is free to use within certain limits. The API supports 38 languages and regions, including Europe, and includes features such as chat functionality and filtering.
During a press briefing, Google Cloud CEO Thomas Kurian highlighted Gemini’s advanced coding skills and sophisticated reasoning, emphasizing that developers can now build applications against it.
The Gemini Pro API in Vertex primarily accepts text input and generates text output. Additionally, the Gemini Pro Vision endpoint, also launched in preview, can process both text and imagery, including photos and videos, producing text output similar to OpenAI’s GPT-4 with Vision model.
Gemini Pro’s integration with Vertex AI allows developers to customize it for specific contexts and use cases using fine-tuning tools similar to those available for other Vertex-hosted models. It can also be connected to external APIs for specific actions and grounded to enhance accuracy and relevance by utilizing third-party data or information from the web and Google Search.
Gemini Pro also benefits from existing Vertex AI capabilities, such as citation checking for fact-checking and improved response quality.
Addressing concerns raised following Gemini’s unveiling, Kurian highlighted control, moderation, and governance options, and to sweeten the deal, Google introduced pricing discounts. The input for Gemini Pro on Vertex AI is priced at $0.00025 per character, and output at $0.00005 per character. These rates are reduced by 4x and 2x, respectively, compared to Gemini Pro’s predecessor. Moreover, Gemini Pro is free to try for Vertex AI customers until early next year.
Google is introducing other features to Vertex AI to compete with rival platforms, such as Bedrock. Some of these features are specifically tailored for Gemini Pro, including its use in custom-built conversational voice and chat agents, as well as search summarization, recommendation, and answer generation features.
Kurian expects Gemini Pro-powered conversational and search features to be available in early 2024. Other additions to Vertex AI include Automatic Side by Side (Auto SxS) for model evaluation, third-party models from Mistral and Meta, and “step-by-step” distillation, creating smaller, specialized, and low-latency models from larger ones. Google is also extending its indemnification policy to include outputs from PaLM 2 and Imagen models.
However, the indemnification policy for the Gemini Pro API will be expanded once the API launches publicly, according to Google.
