I’ve been using gpt-3.5-turbo-16k for a very long time, but don’t seem to find it listed anywhere under OpenAI models. Is it also being deprecated along with:
- gpt-3.5-turbo-0301
- gpt-3.5-turbo-0613
- gpt-3.5-turbo-16k-0613
Here is the relevant link from the deprecations page:
https://platform.openai.com/docs/deprecations/2023-11-06-chat-model-updates
The following model versions remain available:
https://platform.openai.com/docs/models/gpt-3-5-turbo
1 Like
Thank you. My conundrum is that gpt-3.5-turbo-16k is not mention as being deprecated OR being available – it is not in either list.
1 Like
I double-checked and couldn’t find any reference to ‘gpt-3.5-turbo-16k’ anywhere. I guess the question is whether it’s pointing to gpt-3.5-turbo-16k-0613 or one of the newer variants.
How can I find out for sure?
gpt-3.5-turbo-16k is just an alias with only one possible destination. There is nothing for it to point to that is not deprecated and wouldn’t be a massive shift in behavior, quality, lineage, and cost.
1 Like
It could simply point to gpt-3.5-turbo-0125 which also has a 16K context window like all newer 3.5 versions do.
1 Like
It could, but then that redirection to a model with a newly-enforced output limit would not satisfy applications that are producing 4k+ responses - using the model’s primary feature that makes “-16k” unique.
Then there is being redirected to a different (you get what you pay for) price that also would be a concern.
I don’t see price as a concern because prices have been steadily dropping since 3.5 Turbo was initially announced.
The question remains whether GPT-3.5-turbo-16k today refers to the 0613 variant or another one. I do expect it’s the second case but will try to get clarification on this issue regardless.
An API call will tell you what is fulfilling the request to the pointer/alias:
{
"id": "chatcmpl-...",
"object": "chat.completion",
"created": 1725693995,
"model": "gpt-3.5-turbo-16k-0613",
"choices": [
{
"index": 0,
"message": {
"role": "assistant",
"content": "8",
"refusal": null
},
"logprobs": {
"content": [
It is disappointing to see it go, but it won’t be a loss like GPT-3 models, because it was already neutered from its original capabilities by the same output curtailment behavior as the normal gpt-3.5-turbo-0613 received, that has the output often wrapping up far before the potential. “Rewrite” 6k becomes 2k.
It also never fulfilled potential - 7k in and 7k out on a “for each sentence of input, create a new sentence of improved quality” task would give you no modification at all, just repetition.
It seems that gpt-4o-2024-08-+ will have to be the replacement if anyone is still finding particular utility only this 3.5 model can make. The manual choice of a new model, and shutoff of the name, is logical.
1 Like
By default the gpt-3.5-turbo models has been have been coming with 16k context since the release of gpt-3.5-turbo-1106 which itself has 16k context length.
2023-11-06: Chat model updates
On November 6th, 2023, we announced the release of an updated GPT-3.5-Turbo model (which now comes by default with 16k context) along with deprecation of
gpt-3.5-turbo-0613andgpt-3.5-turbo-16k-0613. As of June 17, 2024, only existing users of these models will be able to continue using them.
The depreciation docs show the shutdown date of the following gpt-3.5-turbo models:
| SHUTDOWN DATE | DEPRECATED MODEL | DEPRECATED MODEL PRICE | RECOMMENDED REPLACEMENT |
|---|---|---|---|
| 2024-09-13 | gpt-3.5-turbo-0613 |
$1.50 / 1M input tokens + $2.00 / 1M output tokens | gpt-3.5-turbo |
| 2024-09-13 | gpt-3.5-turbo-16k-0613 |
$3.00 / 1M input tokens + $4.00 / 1M output tokens | gpt-3.5-turbo |
It’s already recommended by by OpenAI to migrate to gpt-4o-mini if you’re still using the gpt-3.5-turbo models, because it’s more economic and powerful.
4 Likes