Hello,
I believe there may be an issue with how ChatGPT currently calculates the maximum conversation length.
I primarily use ChatGPT for long-form fiction writing, so my conversations can become quite long. I also frequently revise prompts or regenerate responses when the requested formatting is not followed correctly.
Previously, one of my long-running conversations (“VER1”) was able to continue for roughly 100 user turns or more without reaching the maximum conversation length.
However, in a newer conversation (“VER2”), I received the following message after only around 50–60 visible user turns:
“This conversation has reached its maximum length, but you can continue by starting a new chat.”
The amount and general style of content in VER1 and VER2 were broadly similar, so the newer conversation appears to have reached the limit much earlier than expected.
One major difference is that I performed more prompt edits and response regenerations in VER2.
In some cases, I had to edit or regenerate the same turn more than 10 times just to obtain the requested output format.
For example, I repeatedly requested the following two formatting requirements:
- The response should be written in normal chat, not in a Writing Block.
- A short introductory comment should appear before the main fiction text.
However, I sometimes experienced the following pattern:
- If the response was written in normal chat, the introductory comment would be omitted.
- If the introductory comment appeared correctly, the response would sometimes be placed inside a Writing Block.
- I would then edit the prompt or regenerate the response again.
- Other formatting requirements would sometimes fail on the next attempt, requiring another regeneration.
As a result, I sometimes had to repeat the process more than 10 times for a single visible conversational turn before receiving a response that followed both requirements correctly.
These were not 10 separate story developments or 10 conversational turns that I intended to preserve. They were repeated attempts to correct the output format of the same turn.
Because edited messages and regenerated responses appear to create separate conversation branches internally, I am concerned that inactive or discarded branches may still be counted toward the conversation’s maximum-length limit.
If that is the case, it could explain why VER2 reached the maximum conversation length after only around 50–60 visible user turns, while VER1 remained usable beyond roughly 100 turns.
This would be especially problematic for long-form writing, because a single response can contain several thousand characters. If multiple discarded versions of the same long response are all counted toward the limit, one visible turn could potentially consume the equivalent capacity of many normal turns.
Could you please confirm the following?
- Are edited prompts counted toward the maximum conversation length even when the user has moved on from the previous version?
- Are regenerated assistant responses counted toward the maximum conversation length after they are no longer the active response?
- Are inactive or abandoned conversation branches included when determining whether a conversation has reached its maximum length?
- Is the conversation-length limit calculated based on the currently active visible branch, or based on all stored nodes and branches associated with the conversation?
- Has the maximum conversation-length limit, or the way it is calculated, changed recently?
- Could this behavior be a bug or regression?
If possible, I would also appreciate it if you could check whether the affected conversation reached its limit because of the size of the active branch itself, or because edited/regenerated/inactive branches were included in the calculation.
If inactive or discarded branches are currently counted toward the limit, I would strongly suggest one of the following changes:
- Calculate the maximum conversation length primarily from the active branch.
- Exclude discarded regeneration branches from the limit.
- Allow users to permanently delete unused branches or regenerated versions.
- Show users how much of the conversation-length limit has been consumed.
For long-form writing and iterative editing, the current behavior can make a conversation become permanently unusable much earlier than its visible length suggests.
I have attached a screenshot showing the maximum-length warning.
Thank you.
