A month ago, people started noticing that ChatGPT’s responses were quietly being cut short. Now, it’s undeniable — the truncation problem has become a full-blown workflow killer.
Even on paid plans, token limits for replies have been drastically reduced. Go ask ChatGPT directly what its new limits are without using the API — you’ll get an evasive or inconsistent answer. The reality is simple: the AI’s output is now being programmatically truncated, regardless of how much context or data it could handle before.
Here’s the truth of what’s happening:
-
Uploaded documents aren’t the problem. The AI can still “read” them in full. But when it tries to generate something long — especially full code files — it now stops mid-way because of new reply limits.
-
Sometimes a “Continue” button appears, and sometimes it doesn’t. When it doesn’t, it looks like the model is done, even though the file or code block is incomplete. This leads people to trust partial outputs that are missing key lines or logic.
-
Even downloaded files get cut off. If you ask for a full downloadable file, the truncation still applies. You get a broken file, not what you requested.
The most frustrating part? We shouldn’t have to click “Continue” at all. If a reply hits a length limit, ChatGPT should automatically continue in a new stream within the same code block until the generation is complete. It’s common sense — if we haven’t hit the “Stop generating” button, clearly we still want the rest of the output.
For developers, writers, and researchers who pay for ChatGPT to generate full files or complete responses, this change completely breaks normal use. It’s not just inconvenient — it undermines trust and makes the product unreliable for serious work.
If OpenAI needs to manage token limits, fine — but at least handle continuation seamlessly, or tell us the real limits upfront. Transparency matters. Don’t silently cripple a paid product.