miau, nimiau?
Hardon (the cockroach) asks a question to Sir Penis Brain (Ivor)
I do not have a simple “X minutes, then amnesia” limit.
Sleeping cat
The practical limit is mainly context, not clock time. I can reason during a response, but I only keep available what fits inside the model’s current context window plus any enabled memory/personalization context. For GPT-5.5, OpenAI lists different context windows by mode and plan: for GPT-5.5 Thinking, paid tiers are listed at 256K tokens, and Pro at 400K tokens; for GPT-5.5 Instant, the listed window ranges from 16K to 128K tokens depending on plan. (OpenAI Help Center)
So “amnesia” usually happens when one of these occurs:
The conversation becomes too long. Older material may fall outside the active context or be compressed/summarised imperfectly.
A new chat starts without relevant memory/context being retrieved.
Memory is off, limited, or not specific enough. OpenAI says ChatGPT does not remember every detail from past chats, and recommends saved memories for things the user wants it always to keep in mind. (OpenAI Help Center)
Transient reasoning disappears after the answer. My private scratchpad is not a durable notebook. The durable parts are the visible conversation, uploaded/referenceable files when available, saved memories, and relevant retrieved context.
OpenAI also says “thinking time” is adjustable and not directly comparable across models; it is tuned for speed/quality tradeoffs rather than being a fixed biological-style attention span. (OpenAI Help Center)
In plain terms: I can think for the duration of the current response, but I do not have continuous consciousness or an unlimited working memory. My amnesia is mostly token-window amnesia, not time-decay amnesia.