Gemini Pro, the multimodal AI model which powers Google Bard, has been made available to developers – both in the Google AI Studio and the enterprise Vertex AI – and Duet AI coding assistance is now generally available.
Google introduced Gemini last week as its latest and most capable AI model, claiming superior performance versus OpenAI’s GPT-4.
Vertex AI is the enterprise option and includes IAM (Identity and Access Management) permissions, the ability to tune Gemini with company data, and stronger data privacy. Vertex also supports other models – including Google’s PaLM, Imagen and Codey, Anthropic Claude 2, and the open source Llama 2.
General availability for Gemini Pro is planned in early 2023, with pricing at $0.00025 per 1K characters or $0.0025 per image for input, and $0.0005 per 1K characters for output.
The Gemini family also includes Gemini Nano, for mobile devices, and Gemini Ultra – currently in private previews – which is a larger model and will be more widely available next year.
Imagen, a text to image technology, has been upgraded to Imagen 2, including copyright indemnification to “give customers peace of mind when using our generative AI products.”
Like Microsoft’s Copilot, Google’s branding for Duet AI is confusing – spanning developer, security and workspace offerings.
Duet AI for developers is Google’s coding assistant. It includes code and chat features, with support for over 20 programming languages and IDE integration for Visual Studio Code, JetBrains IntelliJ, and its own Cloud Shell Editor. Another feature of Duet AI is that Google has worked with “code-assist” and “knowledge base” partners to improve its support for third-party APIs and products, including those from Elastic, HashiCorp, Neo4J, Redis, MongoDB, Atlassian, DataDog, Snyk, LangChain and others. The Big G promises to incorporate Gemini into Duet AI in the coming weeks, but has declared Duet AI for developers production-ready.
“It can complete the code you’re writing, it can generate blocks of code, it can take code in one language, break it up into smaller modules, convert it to another language. It can generate documentation and it can write tests. It can also deploy applications and auto-scale them for you,” enthused Google Cloud CEO Thomas Kurian in a press briefing.
Duet AI in Security Operations is also now generally available. An important feature – the ability to get code suggestions based on internal organization code – is currently in testing and “will be available in Q2 of 2024,” according to today’s post.
Duet AI for Workspace is an assistant for productivity tools like Gmail, Meet and Google Docs, and will also be moved to Gemini in early 2024.
Like Amazon Web Services, Google is coming from behind with its coding assistant. Looking at the VS Code marketplace, GitHub Copilot has over 11 million downloads. Duet AI is packaged with other GCP (Google Cloud Platform) tools as the Google Cloud Code extension, which has just under one million installs – many of which are likely not for the Duet AI features.
Following the AI-dominated AWS re:Invent early this month, and Microsoft’s unrelenting introduction of Copilot-related services at its recent Ignite event, Google seems equally keen to promote its AI offerings.
One aspect of the launch picked up by developers though is the generous free quota, which Google said was 20 times more than its competition. “I wasn’t going to bother even testing Google’s AI products … but with 60 free queries per minute? That’s worth exploring even if only to find out shortly that it’s not worth paying for,’ wrote one developer.
That sounds like a good start, but some have found glitches with the initial launch – receiving messages stating that an “internal error has occurred” and “access restricted” when trying to use AI Studio.