OpenAI used its first developer event to announce new model capabilities and tools for people building with its products. The updates span a larger-context version of GPT-4, user-created GPTs, APIs for assistants and audio, and a commitment to cover certain legal costs.
A larger context window for GPT-4 Turbo
GPT-4 Turbo is an improved version of GPT-4, offered in a text-only option and another that can understand both text and images. OpenAI said its context window is 128,000 tokens, or roughly 100,000 words—four times GPT-4’s context window. Its knowledge cutoff is April 2023, compared with GPT-4’s September 2021 cutoff.
The company gave separate prices for text and image processing. The text-only model costs $0.01/1,000 input tokens and $0.03/1,000 output tokens. Processing a 1080×1080 pixels image with GPT-4 Turbo costs $0.00765.
A larger context window lets a model consider more material within a single interaction. For developers, that could make it easier to build experiences around longer inputs. The announcement also describes an image-capable version, extending the model’s input beyond text.
Users can create and share custom GPTs
OpenAI said users would be able to make their own versions of GPT using prompts, without coding. The company described potential uses ranging from fun to productivity. Enterprise customers will also be able to create GPTs for internal use, drawing on their organization’s knowledge base.
Developers can connect GPTs to external sources such as databases or knowledge bases, including emails. That allows a custom GPT to work with information beyond what a user provides in a prompt, subject to the connected source and the way the GPT is built.
The company plans to let users publish GPTs in a GPT Store, which was described as coming later this month. The store was initially expected to feature creations from “verified builders.” CEO Sam Altman also discussed paying people whose GPTs become popular. ChatGPT Plus and ChatGPT Enterprise customers were to receive access to example GPTs from that day.
New APIs for assistants, images, and speech
The Assistants API is intended to help developers build what OpenAI called “agent-like experiences.” These assistants can retrieve outside knowledge or call programming functions to take a specific action. The examples mentioned at the event ranged from a coding assistant to an AI-powered vacation planner.
OpenAI also made its DALL-E 3 text-to-image model available through an API with built-in moderation tools. The stated output sizes range from 1024×1024 to 1792×1024 in different formats, and each generated image costs $0.04.
For speech, the company introduced the Audio API, a text-to-speech service with six preset voices: Alloy, Echo, Fable, Onyx, Nova and Shimer. The API provides access to two generative AI models and costs $0.015 per 1,000 input characters. Together, these interfaces give developers additional ways to build products around text, images, and spoken output.
Usage figures and a copyright commitment
OpenAI said more than 100 million people use ChatGPT weekly, and more than 2 million developers are building solutions through its API. The company also noted that ChatGPT reached the 100 million monthly user mark a few months after launch, describing it as one of the fastest consumer products to do so.
Another announcement was Copyright Shield, a program concerning legal costs for some customers facing intellectual property lawsuits. OpenAI said it would pay legal fees for customers using its “generally available” developer platform and ChatGPT Enterprise when claims concern content created by OpenAI’s tools.
The commitment was framed around specified products and circumstances. It gives eligible businesses a stated form of legal-cost support, while the event announcement did not describe the program as covering every customer or every type of claim.