Google Adds New Gemini Models to Vertex AI

Google Cloud made a flurry of AI announcements today, with new models available in Vertex AI, upgrades to the Gemini API and new languages in Google Translate enabled by AI. Developers can now take advantage of the 2 million token context window in Gemini 1.5 Pro without needing to be patient on a waitlist. Plus, you can now apply to be one of a limited number of users of Google’s newest image generator, Imagen 3, which can create photorealistic images for marketing or corporate presentations.

Google’s newest AI are open for business

New or higher-performance models of several Google AI are in wider availability in the Vertex AI platform today:

• Gemini 1.5 Flash, a relatively compact model with a 1 million-token context window, is generally available.
• Gemini 1.5 Pro is generally available.
• Imagen 3 is in preview. Apply here.

“Gemini 1.5 Flash makes it easier for us to continue our scale-out phase of applying generative AI in high-volume tasks without the trade-offs on quality of the output or context window, even for multimodal use cases,” said JC Escalante, global head of generative AI at market research firm Ipsos, in a Google press release.

Vertex AI now offers or will soon offer:

• The lightweight Gemini variant Gemma 2: Generally available on Vertex AI next month in two sizes: 9 billion parameters and 27 billion parameters
• Anthropic’s Claude 3.5 Sonnet, which is available now.
• Context caching, a technique used to create higher speed and lower cost for AI requests using repetitive content, is now in public preview for Gemini 1.5 Pro and Flash.
• Provisioned throughput, a feature of Vertex AI for provisioned workloads on Gemini models, is generally available now for users on the allowlist.
• Grounding for better accuracy, in which the AI can check its information against Google Search, is available now. Grounding from third parties such as Thomson Reuters is expected to roll out starting next quarter.
• Grounding with High Fidelity mode, which combines Gemini 1.5 Flash with company data, is now in experimental preview.

Vertex AI is available in a wide variety of geographical regions.

SEE: Here are five ways to search the web with generative AI.

Gemini API can now run code execution and more

Code execution is now possible in Gemini 1.5 Pro and Gemini 1.5 Flash, letting developers run Python within the model and experiment with letting the generative AI iterate and learn from the code. It can be accessed through the Gemini API or Google AI Studio.

In addition, users of the Gemini API can now:

• Use the full 2 million token window on Gemini 1.5 Pro.
• Use context caching for both Gemini 1.5 Pro and 1.5 Flash.
• Experiment with Gemma 2 in Google AI Studio.

Cantonese and 109 other languages added to Google Translate

Google has used the PaLM 2 language model to add 110 languages to the public Google Translate service; this is its largest-ever expansion of this service. A highlight is Cantonese, a language that Google has found difficult to find data in order to add it to Translate in the past because it “often overlaps with Mandarin in writing.”

PaLM 2 has enabled Google to more efficiently add more languages that are similar to each other, Google Senior Software Engineer Isaac Caswell said in a press release about this Google Translate expansion.