I’m experiencing the exact same issue as Athelaperk, and it started a few days ago for me as well.
A number of my long conversations (especially ones that had reached max length) are now failing to load in the web UI with the “Unable to load conversation” error. These same conversations were accessible without issue until very recently, and I regularly relied on them for reference.
A few key observations from my side:
- The issue seems to affect long / maxed-out conversations specifically. Shorter conversations still load fine.
(Not sure if model (e.g. GPT-4o) plays a role for me, as mentioned by others.)
- It reproduces across multiple devices (phone, macOS, Windows) and multiple browsers (Chrome, Firefox, incognito, etc.), so it doesn’t appear to be client-side.
- These same conversations still load in the ChatGPT mobile app, which suggests the data itself is not corrupted.
However, the app is not a real workaround in my case, since it does not allow access to message variants (version arrows), which are only available in the web UI.
- No clear incident on the OpenAI Status page directly linked to this, although recently reported incidents (marked as “resolved”) could be indirectly related.
- Multiple users are reporting this here and on Reddit around the same timeframe.
A few key technical details:
While attempting to load a failing conversation with DevTools (Network tab), I noticed that the following requests consistently fail:
- {conversation-ID}:
/backend-api/conversation/{conversation-id}
- “init”:
/backend-api/conversation/init
With:
- Status Code: 413 (“Content Too Large”)
- Response body
"code": "conversation_length_exceeds_limit"
- Response body
"message": "The conversation is too long, please start a new one."
This suggests that the web UI (or backend) might now be enforcing a stricter size limit when loading conversations.
What’s confusing:
- These conversations were accessible for months
- There was no warning that they were approaching any kind of “unloadable” state
- And now they suddenly can’t be opened at all via web
While I understand that “maxed out” conversations are too long to continue, not being able to load and consult them at all feels unintended—especially given that this worked until recently.
Troubleshooting:
I’ve tried all the usual steps (different devices/browsers, cache/site data clearing, hard refresh, disabling extensions, incognito, log out/in, different network, etc…), with no success.
OpenAI Support response:
I contacted support and shared all the above. An AI support agent responded, suggesting additional standard troubleshooting (none of which worked) and requesting more specific data (including HAR/log data).
The AI agent concluded that it was most likely a server-side failure (due to limitation or regression) affecting the loading of long threads via the web UI, rather than something that can be fixed client-side. It also indicated that the web UI may now be hitting/enforcing a hard limit when trying to load conversations that exceed a certain size—effectively blocking them from loading entirely, not just from being continued (which is why it worked before and now fails across devices).
At this point, the AI support agent proceeded (as is often the case when discussing non-standard issues with AI agents) to:
- suggest standard “workarounds” (data export or using the app)
- mildly imply that the direct cause of the issue was conversations’ length and that there was nothing to be done (e.g. “original thread may remain unloadable on web once it exceeds the limit”; “only mitigations are to collect diagnostics for investigation and to move work into a new chat when a conversation thread is long”).
I had to be quite blunt and remind it that:
- while “maxed out” conversations are too long to continue, they should still be able to load and be consulted
- this issue results in a loss of access to data that was, until recently, accessible
- I am not the only one encountering this issue
- the suggested workarounds don’t fully solve the problem, since neither preserves access to message variants (only available via web UI)
I also highlighted an additional concern: avoiding this in future by starting new conversations earlier is not very practical, as there is currently no visible indication or warning that a conversation is approaching any size or loading limit. The only message users receive is the standard “You’ve reached max length…” notice when attempting to continue the conversation, at which point—based on this behavior—the thread may already be at risk of becoming difficult or impossible to load via Web UI. This makes it difficult to anticipate when to branch off (or continue elsewhere), and increases likelihood of unintentionally reaching this state.
Eventually, the AI support agent acknowledged that the standard workarounds wouldn’t solve my issue and offered to route me to a support specialist.
A case was opened by a support specialist (who requested a screen recording of the error as it occurs), before being escalated further to a specialist team. I am currently waiting for further updates.
Open questions:
While waiting for them to get back to me, I’m also wondering what might be the cause behind this issue/regression to start with.
More specifically:
- Has something recently changed in how large conversations are handled on the backend?
- Is this an intentional limit now being enforced on loading, not just continuation?
- Could this be related to an unintentional regression affecting request size limits (e.g. 413 via Cloudflare (note: server of elements affected by DevTools 413 error seems to be Cloudflare) or similar)?
At the moment, this issue effectively means losing access to parts of past work that were previously available via the web UI.
Would be really helpful to hear if anyone has additional insights or if this is being looked into.
Glad to see I’m not the only one encountering this—hopefully that helps get it on the radar.
Fingers crossed for a fix soon.