Google Expands Access to Gemini Models for Developers

Google is broadening its suite of Gemini large language models available to developers on the Vertex AI platform today.

Gemini 1.0 Pro (previously recognized as Gemini Pro 1.0 just last week—Google has a knack for rebranding) is now officially available after a successful public preview. On a related note, Gemini 1.0 Ultra (also once known as Gemini Ultra 1.0) is now accessible “via allowlist,” which seems to deviate from the standard definition of general availability.

In addition, Google has unveiled Gemini 1.5 Pro (and not Gemini Pro 1.5, naturally), an enhancement to the existing Gemini Pro model that reportedly performs at par with Gemini 1.0 Ultra, the current flagship model. Notably, this model supports a context window of one million tokens, equivalent to about one hour of video, 30,000 lines of code, or over 700,000 words. This model employs a "new Mixture-of-Experts approach," which is currently available in private preview.

Google is also introducing adapter-based tuning within the Vertex platform. This update will soon include support for techniques such as reinforcement learning from human feedback and distillation. Additionally, developers can more easily incorporate real-time data and complex workflows into their models and connect the Gemini model to external APIs through function calling.

For other developer tools, Google highlights that the Gemini API is now accessible via the Dart SDK, enabling seamless integration into Dart and Flutter applications. Moreover, Google is simplifying the Gemini API's usage with Project IDX, its experimental web-based integrated development environment, and is adding integration for Firebase, its mobile development platform, in the form of an extension.

AI, developers, Google, Gemini, Vertex AI