I have a problem with the new context window capacity

ChatGPT
1

I use chatgpt for long term creative writing and most specifically an ongoing fictional storyline that I’ve spent months creating. These stories involve character descriptions, different settings and context and a lot of background.

Recently I’ve been repeatedly getting the context window full warning, and having to start new conversations.

This is really annoying because starting a new chat means I have to transfer a lot of information. Even when I try to continue in a new chat, I still have the same problem.

Ive seen that other people have complained about this and I’d like to know if you are investigating this and are there plans to improve long-term conversation continuity, particularly for creative writing projects?

I really like using chatgpt but constantly having to restart conversations is exhausting.

I’d really like for it to go back to the way it was, back when i could send many messages without having to make a new conversation after sending 20 messages.

2

I can relate to your experience, particularly when it comes to long-term creative writing.

I’ve encountered similar problems while developing one of my own fictional works, part of a much larger narrative universe built over time.

Since October 7, I’ve observed a significant deterioration in the reliability of long conversations and their continuity across sessions.

And the problem goes far beyond simply reaching a context window limit.

A narrative project is not just a collection of plot summaries.

It accumulates established character development, subtle narrative decisions, relationships, chronology, artistic direction, rejected ideas, and details that may only become significant much later.

I’ve experienced context recovery failures several times during conversation resynchronization. The overall story might be recovered, while important details, previously established constraints, and validated decisions are lost, distorted, or incorrectly reconstructed.

Perhaps the richness and complexity of the fictional universe make these transitions more challenging. But that’s precisely why reliable continuity mechanisms matter.

We’ve also documented instances of model switching within an ongoing conversation.

This introduces another important dimension to the problem. Context compaction, resynchronization, and model switching are distinct mechanisms, but their effects on continuity deserve to be examined together.

Even with documented sources and previously validated decisions, I’ve repeatedly had to correct inconsistencies and restore information that was already established.

And sometimes these problems occur within the same conversation, without any interruption or resynchronization.

The result is that instead of progressing with the writing, we spend an increasing amount of time reconstructing what has already been written and validated.

I understand that context windows have technical limits. But a reliable continuation mechanism should preserve more than a simplified summary of the previous conversation.

It should preserve the operational consequences of established decisions, including what has been validated, rejected, corrected, or left unresolved.

Otherwise, starting a new conversation isn’t really continuing the same creative work.

It’s rebuilding it, sometimes imperfectly, at the expense of the author.

I hope OpenAI takes these experiences seriously, because for long-term creative projects, continuity isn’t an optional convenience. It’s a fundamental requirement.

3

You and me both, my friend. You and me both.

4

Hi skye, welcome to the community!

I understand why this is particularly disruptive for long-running creative projects. A fictional universe built over months cannot always be faithfully reconstructed from a short conversation summary.

I looked into OpenAI’s official documentation, and there are two points that may help.

First, diagnosing the context issue:

OpenAI’s Free Tier FAQ acknowledges that automatic continuation summaries may omit earlier instructions or exact wording. It also recommends starting a completely fresh conversation from the sidebar if a continued chat reaches its limit unusually quickly.

If the same warning appears after very few short messages in a genuinely fresh conversation, I would report that separately, including the model, plan, approximate message count, timestamps, and conversation URL.

Second, preserving creative continuity:

One practical workaround is to use a ChatGPT Project with a dedicated Story Bible containing established characters, relationships, chronology, narrative constraints, approved decisions, and rejected ideas.

This won’t eliminate context limits, but it provides a more stable reference than relying entirely on automatic summaries.

I also think the earlier reply raises an important point: continuation should preserve established decisions, not merely summarize previous conversations.

A dedicated Continuity Checkpoint feature that preserves validated facts, corrections, unresolved threads, and narrative constraints could make a meaningful difference for long-term creative work.

Official references:

I hope these suggestions are useful while the underlying issue remains unresolved.

5

Hi my biggest problem is that I am having to create new conversations after about 20 or 30 messages so I need to make summaries of recent context every 20 minutes which is a huge pain. I don’t know if it was a bug or if its a new update but i preferred it much kore before it changed where I could go konths without making summaries for new conversations. I already have major information like character descriptions in the memory to be referred to all the time, I am most concerned about being half way through a secen when i get the notificationthat the context window is full and then having to create a new conversation and give a whole description of recent events just to continue that scene.

6

Thanks for clarifying, Skye! I can understand how frustrating it must be to have your creative flow interrupted in the middle of a scene.

May I ask which ChatGPT plan you’re currently using?

If you’re on the Free plan, you might consider trying ChatGPT Go. It’s a relatively affordable upgrade that offers higher usage limits and additional features compared to the Free tier.

Of course, I can’t say whether upgrading would resolve the specific issue you’re experiencing, but knowing your current plan would help us better understand the situation.

I hope you find a solution that lets you enjoy writing your stories without these constant interruptions!

7

I’m currently using the free plan and I don’t intend to upgrade to chatgpt go or plus since I’ve seen that other people use the upgraded plans and struggle with the same problem. I don’t want to pay for a different plan if I’m going to have the same issue.

8

That makes sense, and I wasn’t suggesting that upgrading would necessarily solve the issue.

I think your concern highlights an important distinction: subscription benefits and the reliability of conversation continuity are not the same thing.

Users should have a clear understanding of what happens to their conversation history, memory, and context when the product changes, regardless of their subscription plan.

Hopefully, OpenAI can provide more transparency around these behaviors so users can make informed decisions.

Thanks for clarifying your perspective!

9

Upgrading and subscribing doesn’t really help, as even payed customers are struggling with this limit.

Buf I applaud the ability to advertise useless information that doesn’t work.