OpenAI's first DevDay ships GPTs and an Assistants API

OpenAI held its first developer conference, DevDay, and used it to announce a preview of GPT-4 Turbo alongside two new ways for outsiders to build on the company’s models: GPTs and an Assistants API. Sam Altman framed the event as OpenAI positioning itself as a platform rather than only a model provider.

GPT-4 Turbo extended the model’s context window to 128,000 tokens — enough, OpenAI said, for more than 300 pages of text in a single prompt — added knowledge of world events up to April 2023, and cut pricing to one cent per thousand input tokens and three cents per thousand output tokens, cheaper than standard GPT-4. It also unified vision, DALL-E 3 image generation and text-to-speech behind one model.

GPTs let anyone create a customised version of ChatGPT — with specific instructions, extra knowledge and selected tools — without writing code, aimed as much at ordinary users as developers. OpenAI said it would launch a GPT Store later that month so builders could publish and, eventually, be paid for widely-used GPTs. The separate Assistants API gave developers persistent, stateful conversation threads together with built-in tools such as Code Interpreter and function calling, addressing problems most developers had previously had to solve themselves.

The keynote was widely read as OpenAI moving to own the application layer built on top of its models, not just the underlying model. That platform push was interrupted eleven days later, when OpenAI’s board removed Altman as chief executive in a move it did not explain in detail; he returned five days after that with a reconstituted board. The GPT Store itself did not open to the public until January 2024, later than DevDay had signalled.