A few weeks ago, OpenAI announced its newest large language model for generative AI services, GPT-4 Turbo, during its first developer event. Then Microsoft announced it would add the GPT-4 Turbo model to its Azure OpenAI Service.
Today, Microsoft announced an enhancement of its Azure OpenAI Service, as the GPT-4 Turbo with Vision model is now available as a public preview version for its customers.
In a blog post, Microsoft stated:
This advanced multimodal AI model retains all the powerful capabilities of GPT-4 Turbowhile introducing the ability to process and analyze image inputs. This provides the opportunity to utilize GPT-4 for a wider range of tasks, including accessibility improvements, visual data interpretation and analysis, and visual question answering (VQA).
In addition, Microsoft has added even more features for Azure OpenAI customers with the GPT-4 Turbo with Vision preview. One of them is Optical Character Recognition (OCR), which will examine images and extract any text in those images so they can be integrated into the user"s prompt.
Another feature that is a part of GPT-4 Turbo with Vision is object grounding, which will let the AI examine an image and reveal its key objects in response to user"s text prompts. The same thing can also be done with the AI analysingspecific frames of a video.
Microsoft added:
By combining GPT-4 Turbo with Vision, Azure AI Search, and Azure AI Vision, images can now be added with text data, utilizing vector search to develop a solution that connects with user’s data, enabling an improved chat experience.
Pricing for the service will be $0.01 per 1000 tokens for Input and $0.03 per 1000 tokens for Ouput, with different pricing for enhanced features.
Currently, GPT-4 Turbo with Vision features can be accessed in the Australia East, Sweden Central, Switzerland North, and West US regions of Azure OpenAI. Customers who are accessing the public preview version of GPT-4 Turbo with Vision will be automatically updated to the "stable, production-ready release in the coming weeks" when it becomes generally available.