DeadAI Dev Tools

GPT-3 davinci (and base model family)

by OpenAI

Announced

2024-01-04

Complete

2024-01-04

Refund window closes

—

Refund status

Refund window CLOSED

OpenAI's original GPT-3 base model family — davinci, curie, babbage, ada — was deprecated and removed from the OpenAI API on January 4, 2024. These were the original "completion" endpoints that powered the first wave of GPT-3 applications between 2020 and 2023. Most developers had migrated to gpt-3.5-turbo or gpt-4 before the deadline, but a long tail of legacy applications silently broke. There is no path to revive a GPT-3 davinci application; the migration target is gpt-4-turbo with prompt rewrites.

Refund flow

1
Not applicable. The OpenAI API is pay-as-you-go — no pre-paid credits or subscriptions tied to specific deprecated models.
2
Costs simply stop accruing for the deprecated endpoints; you don't owe anything for unmade API calls.

Migration path

Recommended successor

gpt-4-turbo or gpt-4o (chat-completion API)

Why this is the closest fit

GPT-4 is the direct successor for general-purpose language tasks. For high-volume / cost-sensitive applications, gpt-3.5-turbo is cheaper and still chat-style.

What differs from the original

GPT-3 davinci was a completion model (single prompt). GPT-4 is chat-completion (structured message arrays). Migration requires restructuring prompts, updating API signatures, re-testing outputs (stylistically different), and accepting 5-30× higher per-token cost.

What this tool meant

GPT-3 davinci was the foundation of the modern generative-AI era. Released as a paid API in June 2020, it powered most high-profile AI applications between 2020 and 2023: Jasper, Copy.ai, Notion AI, Sudowrite, hundreds of GPT-3 wrappers and chatbots, and the original ChatGPT (a fine-tune of GPT-3.5). Its deprecation closed a chapter of AI history. The completion-style API was harder to use safely (no built-in conversation structure, more prompt-injection surface area) but it was the model that proved generative AI's commercial viability. For developers nostalgic for the GPT-3 era, open-source ecosystems (LLaMA, Mistral, Mixtral) preserve the completion-style API pattern.