Update: the latest GPT-3.5 Turbo model, gpt-3.5-turbo-0125, now has lower prices, higher accuracy at responding in requested formats and includes a bug fix for a text encoding issue for non-English language function calls. Read more.
What’s new?
GPT-3.5 has been updated to feature higher accuracy at responding in requested formats, as well as a bug fix which caused a text encoding issue for non-English language function calls. The prices for this model have been reduced by 50% to $0.50 / 1 million tokens (or $0.0005 / 1K tokens) and output prices are reduced by 25% to $1.50 / 1 million tokens ($0.0015 / 1K tokens).
The model can be accessed by passing gpt-3.5-turbo-0125 as the model parameter in the API. It is available with a 16K context window by default and continues to support more reliable JSON outputs.
Does GPT-3.5 have an updated knowledge cutoff?
The knowledge cutoff for GPT-3.5 remains September 2021.
How is this model different from GPT-4 and GPT-4 Turbo?
Our GPT-3.5 Turbo series is still faster than GPT-4 and GPT-4 Turbo. While GPT-3.5 is a highly capable model, we recommend using GPT-4 for improved reliability and creativity on higher complexity tasks. Fine-tuning GPT-4 is only available through an experimental program today, whereas GPT-3.5 fine-tuning is generally available. You can see which models are available to be fine-tuned in our fine-tuning guide.