ChatGPT conversations are reaching maximum length much sooner than before

ChatGPT Bugs
1

Environment

  • Surface: Web / Windows 11
  • Browser or app version: Google Chrome 154
  • Affected feature and model, if relevant: Long conversations / maximum conversation length / GPT-5.6 Sol
  • Account plan: ChatGPT Plus

Bug

Long conversations are reaching the maximum conversation length significantly sooner than they did previously.

I use ChatGPT Plus primarily for long-form fiction writing, keeping one project in a single continuous conversation for continuity. In the past, conversations with similar message lengths and usage patterns could continue substantially longer, often long enough for me to complete the project within one chat.

Recently, however, new conversations have been reaching the maximum-length limit after only approximately 50–60% of the amount of content I was previously able to use.

I expected the practical usable conversation length to remain comparable to what was previously available. Instead, the conversation becomes locked much earlier and I am forced to start a new chat.

This is important because starting a new conversation and relying on Memory or previous-chat references is not equivalent for long-form fiction. Precise continuity depends on the full conversation history.

I would like to know whether this is an intentional reduction in the maximum conversation length, a change in how conversation size is calculated, or a regression/bug.

Reproduction

  1. Start a new ChatGPT conversation.
  2. Use it continuously for long-form writing over many turns, with responses typically several thousand characters long.
  3. Continue the same project in the same conversation.
  4. Eventually the conversation displays the maximum-length message and prevents further continuation.
  5. Compared with older conversations created and used in essentially the same way, the limit is now reached at roughly 50–60% of the previously usable conversation length.

This is not easily reproducible with a single minimal prompt because the issue concerns the cumulative usable capacity of a conversation.

Diagnostics

Exact error

This conversation has reached its maximum length, but you can keep talking by starting a new chat.

When it happened (with timezone) and how often

Most recently on October 4, 2026, KST (UTC+9).

I have encountered this in recent long-running conversations, and the practical conversation limit appears consistently much shorter than it was in older conversations with similar usage patterns. This does not appear to be an isolated one-time error.

I still have access to both older long-running conversations and more recent conversations affected by this issue, and I can provide additional details or screenshots if needed.

2

I’m seeing essentially the same problem on ChatGPT Plus, and I’ve already gone through a fairly extensive support escalation about it.

In my case, this is not just one conversation reaching its normal context limit.

Over roughly the last couple of weeks I have repeatedly experienced:

  • conversations reaching **“maximum conversation length” much earlier than comparable conversations used to;
  • loss or distortion of information from the same active conversation;
  • ChatGPT forgetting things established only a few messages earlier;
  • Error in message stream / missing responses;
  • interrupted connections and responses failing to complete;
  • inconsistent model/mode behavior;
  • conversations behaving differently after page refreshes;
  • problems continuing long analytical workflows that previously worked normally.

I use ChatGPT heavily for professional analytical work, so I eventually opened a support case.

I provided OpenAI Support with:

screenshots of the premature maximum-length messages, exact timestamps, a direct URL to one affected conversation, details about Project usage/files/reasoning mode, and descriptions of the other context and message-stream failures.

Support acknowledged that my complaint was specifically about conversations reaching their limits earlier than expected, rather than simply the existence of context limits.

However, I never received an answer to the most important technical question:

Was the actual context consumption of the affected conversation normal, or had something changed/regressed?

Instead, support explained that capacity may vary depending on model, reasoning level, context, files, task complexity, Fast mode, and other factors.

What particularly concerned me was this part of their response:

“If you require additional usage beyond what is included with your current subscription, you can review the higher-tier plans and available usage options shown for your account to determine whether another plan better meets your usage needs.”

This was in response to a complaint that functionality I had already been using successfully on Plus appeared to have degraded.

I then asked explicitly whether the specific conversation URL I supplied had actually been technically reviewed and whether its context consumption had been determined to be normal or abnormal.

Support requested a screen recording as an additional requirement, despite already having screenshots, timestamps, the conversation URL and a detailed description of the problem.

Because the affected conversations contain confidential/proprietary professional material, producing a recording while safely redacting everything would require even more of my working time. I declined to continue doing additional unpaid diagnostic work after already documenting the issue extensively.

OpenAI ultimately said they would pass the feedback internally, but would not continue the specific investigation without the recording.

I also requested compensation because the degradation had repeatedly interrupted paid professional use of ChatGPT. That request was explicitly denied — no complimentary service, subscription credits, increased limits or other compensation.

So at this point I have two separate conclusions:

What I can actually demonstrate:
The effective usable length/continuity of my ChatGPT conversations has degraded compared with my previous experience, other users are independently reporting very similar behavior, and OpenAI Support did not provide a technical explanation for why this changed.

What I suspect, but cannot prove:
I am concerned by the fact that users are reporting reduced practical capacity while Support’s proposed solution includes considering a more expensive subscription tier.

I am not claiming that deliberate degradation for upselling has been proven. I have no internal OpenAI evidence that would establish intent.

But from a customer perspective, the sequence is difficult to ignore:

previous workload worked → practical capacity appears reduced → users report the same thing → no clear technical explanation → higher-tier plans are suggested as a solution.

If anyone else experiencing this has comparable old/new conversations, it would be useful to document:

  • plan/model;
  • approximate date the behavior changed;
  • whether the chat is inside a Project;
  • whether files are involved;
  • how much shorter new conversations appear to be compared with similar older ones;
  • exact wording of the limit/error message.

The more independently reproducible examples there are, the harder it becomes to dismiss this as a single overloaded conversation or individual account issue.