If your production app still runs on a deprecated OpenAI Completions model, you have until early 2024 before it breaks.

Thousands of apps are quietly running on models OpenAI is about to kill

Teams that built on the older Completions API endpoints never had a forcing function to migrate. Now they do, and the clock is already running.

GPT-4 access is no longer a waitlist problem

GPT is now open to every developer with an API key, no application required. You call the Chat Completions endpoint, pass your messages array, and get GPT-4 output back at scale. OpenAI also moved GPT-3.5 Turbo, DALL·E, and Whisper into general availability in the same release.

The teams who feel this migration deadline first

• Backend engineers maintaining production chatbots built on text-davinci-003 who need a migration path before services cut off
• Product managers at SaaS companies who approved GPT-3-era integrations and now own the upgrade decision
• AI consultants billing clients on Completions-based pipelines who need to reprice and rearchitect before January

The deprecation is not a soft sunset. OpenAI has named a hard retirement window, which means any team that delays is building toward a guaranteed outage.

The GPT-4 waitlist held back an entire tier of products for over a year

Anthropic shipped Claude API access broadly while OpenAI kept GPT-4 gated, and a measurable slice of developer mindshare shifted during that window. General availability closes that gap, but the teams that already migrated their architecture to alternatives will not automatically return.

What you can do with this right now

• Swap Completions API calls to Chat Completions format before the retirement date
• Run GPT-4 in production without requesting special access or joining a waitlist
• Call DALL·E and Whisper endpoints under the same generally available terms
• Audit existing API usage logs to find which deprecated model IDs still appear

Pricing runs on per-token rates published on OpenAI’s platform page. GPT-4 input tokens cost more than GPT-3.5 Turbo, so a cost audit before migrating is worth the hour.

One real limit teams keep hitting

General availability does not mean unlimited capacity — rate limits still vary by account tier, and high-volume production workloads can still hit throttling walls during peak periods.

Anthropic and Cohere both want this migration traffic

Anthropic’s Claude 2 API targets the same professional developer segment with a longer context window. Cohere positions its Command models specifically for enterprise teams nervous about vendor concentration.

The Completions API era is ending and most teams are not ready

The retirement of older OpenAI models is the kind of infrastructure deadline that gets ignored until it is two weeks away. We cover tools like this every Friday — subscribe here and we’ll send the best ones straight to you.