OpenAI has announced the launch of GPT-4 Turbo model during the sidelines of DevDay. The company claims that the new GPT-4 Turbo is more capable, cheaper, and provides support for a 128K context window. The company has announced new additions and enhancements with reduced pricing across various platforms. OpenAI also released new Assistants API using which developers can build their own assistive AI apps. These apps will have goals and can call models and tools. The platform also added new multimodal capabilities in the platform such as vision, DALL E3 image creation, and TTS capabilities. The new features will be made available to Open AI customers starting November 7.
GPT-4 Turbo Features
OpenAI has launched the next-generation GPT-4 Turbo, which is more capable and delivers information about world events up to April 2023. The platform features a 129K context window, which is equivalent to more than 300 pages of text in a single prompt. OpenAI also optimized the performance and hence the company is able to provide GPT-4 Turbo at 3x cheaper price for input tokens and a 2x cheaper price for output tokens. The newly launched GPT-4 Turbo is currently available for all paying developers by passing gpt-4-1106-preview in the API. The company is gearing up to launch the stable production-ready model by the end of 2023.
With the help of GPT-4 Turbo, you will be able to call multiple functions in a single message. You just need to send one message to request multiple actions instead of multiple roundtrips within the model. The company has improved function calling accuracy with the new model likely to return the correct function parameters.
JSON Support
According to OpenAI, the GPT-4 Turbo is capable of performing better than the previous models on advanced prompts. The model also provides support for new JSON mode. Moreover, the response_format API parameter enables the model to constrain its output to generate a syntactically correct JSON object. The JSON mode will be useful for developers who make use of JSON in the Chat Completions API outside of function calling. The new seed parameter enables reproducible outputs, which will be useful for use cases such as replaying debugging requests, writing more comprehensive unit tests, and much more. OpenAI also launched the log probabilities for the output tokens generated by GPT-4 Turbo and GPT 3.5 Turbo models.
GPT-4 Turbo Accessibility Friendly
The newly introduced GPT-4 Turbo will accept images as inputs in the Chat Completions API. This feature will help in several scenarios such as captions generation, analyzing real-world images in detail, and reading documents with figures. BeMyEyes makes use of this technology to assist blind people or those with low vision with daily tasks. As a developer, you can access this functionality with the help of gpt-4-vision-preview in the API.
Imaging & Voice
You can integrate DALL E3 directly into the apps and products via Images API by specifying dall-e-3 as the model. You should note that top companies like Snap, Shutterstock, and Coca-Cola have employed DALL E3 to dynamically generate images and designs for customers and campaigns.
With the help of GPT-4 Turbo, you will be able to generate human-quality speech via the TTS API. The model provides 6 preset voices with dual variants such as tts-1 and tts-1 hd models. While the tts is optimized for real-time use cases, the tts-1-hd is optimized for quality.
