OpenAI yesterday made a surprising announcement without much fanfare. OpenAI"s flagship frontier model GPT-4o now supports a much larger number of output tokens.
The experimental version of GPT-4o, named GPT-4o Long Output, can now support a maximum of 64,000 output tokens per request. This will help developers use the GPT-4o model for use cases that require longer completions. Developers who are part of OpenAI"s alpha program can access GPT-4o Long Output by using the "gpt-4o-64k-output-alpha" model name.
Since long completions are more costly, OpenAI has increased the per-token pricing of this model. The gpt-4o-64k-output-alpha will cost $6.00 / 1 million input tokens and $18.00 / 1 million output tokens.
On a related note, Microsoft yesterday announced the release of the GPT-4o mini API with vision support for Global and East US Regional Standard Deployments. Microsoft will bring it to more regions soon.
With GPT-4o mini vision support, developers can use the APIs to process images and videos in apps involving visual recognition, scene understanding, and more. Also, GPT-4o mini can generate detailed and contextually accurate text outputs from visual inputs.
The Microsoft Azure team wrote the following regarding GPT-4o mini"s availability on Azure:
"With the addition of vision input capabilities, GPT-4o mini expands its versatility and opens new horizons for developers and businesses. This enhancement allows users to process and analyze visual data, extracting valuable insights and generating comprehensive text outputs."
Azure customers can now experiment with GPT-4o mini in the Azure OpenAI Studio Playground and they can learn more about the model here.
These advancements in OpenAI"s and Microsoft"s offerings demonstrate the rapid progress in AI landscape, particularly in language and vision models. As these models become more powerful and accessible, they open up new possibilities for developers and businesses to rethink the experiences they offer for consumers.