OpenAI’s new GPT 3.5 Instruct model announced through E-Mail, pasted below. This is a drop-in replacement to text-davinci-003. It is widely used in code snippets around the web, so it’s good to finally have an optional available that combines the traditional TextCompletion interface and mode of operation with the benefits of GPT-3.5-turbo. Noteworthy is that log probabilities are also supported (see screenshot). Commenters have speculated that gpt-3.5-turbo was based on the same MoE architecture as GPT-4 (meaning: actually several models behind the scenes), so logprobs would simply not be possible to be had.

Caution: GPT-4 is still safer to operate, in multiple ways.

We are excited to announce the release of gpt-3.5-turbo-instruct, our latest model that serves as a replacement for several deprecated models, such as text-davinci-003.

Key Features: Gpt-3.5-turbo-instruct is an InstructGPT 3.5 class model. It’s trained similarly to previous Instruct models such as the text-davinci series while maintaining the same speed as our turbo models.

Pricing: We are committed to making cutting-edge technology accessible, so we have priced gpt-3.5-turbo-instruct in line with our other turbo GPT-3.5 models with 4K context.

Thank you for being a part of our journey and for building with OpenAI. Your support enables us to continually advance and bring you the best in AI technology.

Edit: (actually, it may not be a 100% drop-in replacement: check that you’re using the appropriate tokenizer. Referring to the Cookbook, for text-davinci-003 it’s “p50k_base”, and for the newer models it “cl100k_base”. This was last updated before this Instruct model release, so best be careful and double-check)