OpenAI announced general availability of the GPT-4 API on 6 July 2023, removing the waitlist that had restricted access since the model's March launch [source]. All existing API developers with a history of successful payments gained immediate access to GPT-4 with 8K context, with plans to open access to new developers by the end of July and then raise availability limits based on compute availability.

The announcement included a significant deprecation timeline. OpenAI stated it would retire older models in the Completions API beginning 4 January 2024, specifically targeting models from the GPT-3 and GPT-3.5 series including ada, babbage, curie, davinci, and text-davinci-003 [source]. Users of these legacy models were directed to migrate to newer replacements: gpt-3.5-turbo for most use cases, or text-embedding-3-small and text-embedding-3-large for embeddings workloads.

The company also announced GPT-4 with 32K context would become available to existing GPT-4 users, and that GPT-3.5 Turbo, DALL·E, and Whisper APIs would follow with general availability. Fine-tuning capabilities for GPT-4 and GPT-3.5 Turbo were promised for later in 2023 [source].

The deprecation represented a substantial shift in OpenAI's API surface. Developers relying on Completions API endpoints faced a mandatory migration to Chat Completions, requiring code changes to accommodate the different request and response formats. OpenAI provided a transition period of approximately six months between announcement and enforcement, though the company retained the ability to adjust the timeline based on compute constraints.

The announcement made no mention of performance parity between deprecated and replacement models for existing workloads.