At today’s Gemini at Work event, companies like PUMA, Snap, and Warner Brothers Discovery are showcasing the powerful shift toward AI-driven production. With 61% of enterprises now deploying generative AI use cases in production, our customers are leading the way with transformative applications. This momentum is evidenced by a 36x surge in Gemini API usage and a nearly 5x increase for Imagen on Vertex AI in 2024. This post will answer essential questions about moving from experimentation to production with generative AI, while introducing several exciting updates to our models and platform.
1. What are the latest Gemini and Imagen model updates for maximizing performance while balancing costs?
Our goal at Google Cloud is to provide enterprise-ready models that excel in performance, latency, and cost-efficiency. Recent updates to our Gemini and Imagen models achieve just that:
New Enhancements to Gemini Models:
Gemini 1.5 Pro and Flash models now deliver significant advancements in math, long-context understanding, and vision processing, with the Gemini 1.5 Flash model achieving nearly 2.5x the speed of GPT-4o mini. A 2M context window for the Gemini 1.5 Pro model is also available, ideal for analyzing extensive data like two-hour videos, large codebases, or lengthy contracts.
Expanded Use Cases:
With these enhancements, industries can unlock advanced applications—healthcare providers can anticipate disease progression, finance companies can optimize investments and detect fraud, and retailers can deliver hyper-personalized recommendations.
Affordable Access with Gemini 1.5 Pro:
Following an 80% cost reduction for Gemini 1.5 Flash in August, the Gemini 1.5 Pro now offers a 50% cost reduction on Vertex AI as of October 7, 2024.
Imagen 3 Model Update:
Now generally available on Vertex AI for select customers, Imagen 3 excels in prompt comprehension, photorealistic quality, and text control within images, fueling diverse applications from advertising to personalized visuals. For instance, retailers and branding agencies are enhancing product imagery, while creative platforms use Imagen 3 for stock photos and design ideas.
Cimpress and L’Oréal offer real-world examples of Imagen 3’s impact: Cimpress sees 130% quality improvements in customized design options, while L’Oréal utilizes Imagen 3 to craft dynamic product visualizations, accelerating creative ideation.
2. How can I ensure my AI outputs are precise and valuable?
Building value with AI requires tools that empower you to control and refine outputs effectively. Vertex AI now includes advanced features to facilitate this:
Controlled Generation: Fine-tune outputs with pre-built formats like JSON or ENUM, ensuring they fit specific applications.
Batch API (in preview): Send multiple multimodal prompts in a single batch request at a 50% cost reduction, ideal for non-latency sensitive tasks like sentiment analysis.
Supervised Fine Tuning (SFT): Available soon for Gemini 1.5 Flash and Pro, SFT enhances model precision for unique enterprise needs.
Prompt Optimization and Management SDK: Optimize prompts effortlessly with the Prompt Optimizer, and manage and version them with the Prompt Management SDK for improved outcomes.
3. How can I deploy and expand my AI initiatives confidently?
Enterprise AI requires reliability and customization. Our latest Vertex AI enhancements address these needs:
Gen AI Evaluation Service: Define custom evaluation criteria to assess how well models align with specific use cases.
Data Residency Options: Expanded data residency commitments now support Canadian, Japanese, and Australian organizations, helping to meet data sovereignty requirements.
Transforming AI Aspirations into Tangible Results
These enhancements respond to customer feedback, reflecting Google Cloud’s commitment to providing an open, flexible AI ecosystem. Whether you’re new to AI or scaling advanced applications, our updates and resources will support your innovation journey from experimentation to impactful production. Explore insights and tools from Gemini at Work to unlock the potential of generative AI with Vertex AI.
