Google is adding Gemini big language models to its Vertex AI platform for developers.
After a week of public preview, Gemini 1.0 Pro (or Gemini Pro 1.0, since Google is wonderful at branding) is now widely accessible. Google claims that Gemini 1.0 Ultra (formerly Gemini Ultra 1.0) is now widely accessible “via allowlist,” which isn’t how general availability works.
Google also introduced Gemini 1.5 Pro (not Gemini Pro 1.5), an upgrade to its Gemini Pro model that performs like its flagship Gemini 1.0 Ultra. This approach can manage one million tokens, which may be more significant. Over 700,000 words, 30,000 lines of code, and one hour of video. Google calls this model a “mixture-of-experts approach” and is under private preview.
Google is introducing adapter-based tuning to Vertex, with reinforcement learning from human input and distillation coming shortly. Additionally, developers may now call functions and add up-to-date data to their models for more complicated processes. This lets developers link Gemini to external APIs.
Google announces that the Dart SDK now provides access to the Gemini API for Dart and Flutter developers. It’s also making it easy for developers to utilize the Gemini API with Project IDX, its experimental web-based integrated development platform, and Firebase, its mobile development platform, using an extension.
