I have 2 actions defined in my Custom GPT. These actions work fine with GPT 5.3.
I have instructions to execute the 2 actions (market data via an API) before building the response.
These actions worked fine in 5.3. 5.4 and 5.5 (all versions) do not call the defined action.
ChatGPT itself reports this:
There's an interesting discrepancy: in this chat I can see tool definitions corresponding to TradingEconomics and the EIA, but the execution environment isn't allowing me to invoke them. That points to a separation between tool definitions being presentandtool execution being available in this environment. It's consistent with a runtime limitation rather than a problem in your GPT's configuration.
From a tooling perspective, why provide tools that are not supportable or break under in new environments? How many enterprise customers use CustomGPT’s? Is that a viable, supportable route?
I see quite a few threads having similar issues. Is there a list of apps/tools that are no longer supported?
Me too. I’m seeing very similar behaviour with a much larger Airtable workflow.
I’m using a Custom GPT (GPT-5.5) with OpenAPI Actions against Airtable. The GPT has successfully completed 44 full production workflows using the same:
OpenAPI schema
Airtable base
PAT authentication
GPT Instructions
Knowledge files
No configuration changes have been made.
Over the last few days the behaviour has changed dramatically. The Actions runtime is now predominantly unavailable, with only brief periods where some Actions execute before failing again.
I’ve observed multiple failure modes, including:
Actions reported as unavailable in a particular chat.
Some Actions available while others (e.g. createLessonRecord) are missing.
The first Action succeeds but subsequent sequential Action calls fail.
An Action reports success but the returned payload is not available to the model.
Uploaded files have intermittently become unreadable (filenames visible, contents unavailable).
The same GPT may work in one conversation and fail in another without any changes to the GPT, schema or Airtable configuration. Sometimes the identical Action works again a few minutes later.
I’ve already tried:
New conversations
Different browsers (Edge and Chrome)
Incognito mode
Hard refresh
Builder Action tests
The behaviour persists.
I’ve reported it to OpenAI Support and they have escalated it to a support specialist. At the moment this is preventing completion of a production workflow because the runtime frequently becomes unavailable before the sequence of Airtable Actions can finish.
Is anyone else seeing similar behaviour with GPT-5.5 or GPT-5.6, particularly with multiple sequential Action calls?
I’ve literally been fighting with this exact issue for the last two weeks now, or honestly maybe the same time because my days are running together with how much debugging I’ve been doing. I also submitted a ticket that was escalated, but haven’t heard anything. It’s got so frustrating that I’ve started building my own system on my website so that I can control the tooling and the system merely uses OpenAI’s API. Now if these issues could be stable for a single day, I’d be able to finish that system for when these… hiccups happen.
Thanks for the detailed reports. Asir’s Thinking workaround is worth testing if available. Please add your GPT link and one recent failing conversation to your existing Support ticket, including the model and time. If another supported model works with the same prompt and unchanged GPT settings, include that conversation too. Please share these privately through Support. Please do not support multiple support cases instead just share 1 case id from 1 affected user here and we can use that to solve the overall issue. Thank you!