I think with over 900 million free users worldwide, OpenAI would not abandon their most important user group. AI literacy must be a key objective for this startup.
I’m facing the same issue. I created a custom GPT that used to call certain actions, but now on my free (Go) account it says that no actions are available. However, when I use my Plus account, it performs the actions normally. This started happening recently.
Just out of curiosity, I thought Gemini Gems don’t have a feature similar to action calling. How were you thinking of implementing your workflow there?
Do you refer to over-refusal with Assistants on the API (indeed where anyone with “users” should be coding their customer-facing product with a “developer” message), or are you talking about ChatGPT “GPT”'s instruction field been seen as untrustworthy, or an actual failure with your “actions” there?
This topic is clearly now about non-paying or “Go” ChatGPT users being unable to use others’ GPTs if with “actions” that connect to the GPT creator’s network-based service. That symptom is not a model behavior.
@OpenAI_Support mentioned that they were looking into the issue, but it hasn’t been resolved and there has been no follow-up since then. Once a commitment is made, I believe it’s important to see it through. Could you please provide an update?
Thanks @OpenAI_Support. The issue has been fixed, however there is another issue.
Earlier, users had option to choose from the legacy models and select model recommended by custom GPT. In 5.1 the option to use previous models is not there with free users. Unfortunately, 5.1 is hallucinating a lot I have just compared results for my custom gpt for 4o (which was the best model) and 5.1, I can clearly see that 5.1 is hallucinating and not following the instructions.
Can you please advise by when you will release option for users for select legacy models. I remember during 5o rollout, legacy model selection was provided a few days later, are you going to do that this time too?
Can you please make this part of your regression testing so that the existing functionality does not break after model upgrade. This is the 2nd instance when after rollout existing custom actions broke for users , this time it has taken more than 5 days which is unacceptable.
Could you create one or two new topics for the suggestions? This thread is focused on the resolved custom actions issue and the ideas might get lost once it’s closed.