[CjHuber]

Interesting to hear people like gpt-5.5. For me it feels smart only at one shot prompts, but if you try to build up session context before doing something it feels magnitudes inferior to Claude. I'm almost sure its because the thinking of previous turns is stripped with the responses API, so if I tell it to analyse something deeply, what remains of the understanding in future turns is only the short response text of that analysis

[porker]

For gpt-5.5 I build up that session context into a markdown file, and then I start a new session and give it the markdown file with the instructions for what to do.

I'm guessing this works better because it can always go back and re-analyze the saved context.

[prodigycorp]

Do you have a source for this? I'm pretty sure responses api is only there to obfuscate reasoning, but that they're still keeping reasoning traces in the backend.

[tedsanders]

I work at OpenAI and can confirm that's correct: reasoning tokens are discarded after each new user turn (though not after each message or tool call).

Our docs show a diagram here:

https://developers.openai.com/api/docs/guides/reasoning

> Input and output tokens from each step are carried over, while reasoning tokens are discarded.

Keeping reasoning tokens around is better for caching and for remembering past insights, so you might reasonably wonder why we designed it this way. The main benefit of dropping reasoning tokens is that you can fit a lot more work inside the model's context window before you're forced into a slow and lossy compaction step. This was a larger consideration with our earlier reasoning models that had shorter context windows (~200k), longer thinking times (up to ~100k per message), and poor compaction. However, now that we've shipped longer context windows, we've trained our models think much more efficiently, and we've made compaction way better than it used to be, the balance of factors is changing. Tune in Thursday!

[CjHuber]

> fit a lot more work inside the model's context window before you're forced into a slow and lossy compaction step

This is something I never understood. Why the reasoning is not included until the context is full, then the reasoning stripped optionally to allow the conversation to continue. and only then when its truly full offer a compaction. Was it to optimize caching? Well I guess it doesn't matter now that you hinted that this choice was made because of prior limitations and may change very soon

[embedding-shape]

> Why the reasoning is not included until the context is full, then the reasoning stripped optionally to allow the conversation to continue. and only then when its truly full offer a compaction.

Models are typically trained (at longer conversations/more turns) either with or without the reasoning still in the conversation. If you train a model with those, then using it without them, the model will perform a lot worse, same vice-versa if you train without but then end up using the model with them.

That's why you'll see some models have it and others don't, and trying to use them another way, will make them worse, they weren't trained like that.

So why aren't the models trained with both? I'm guessing that sort of permutation in the training would lead to double the amount of training time being needed, as you know effectively will have two variants of every session you train on, with and without the reasoning.

[throwuxiytayq]

dude what the hell

[throwuxiytayq]

> thinking of previous turns is stripped with the responses API

Why do they store an encrypted reasoning payload in the session file and pass it to the API? Just a ruse? Reasoning isn’t even that many tokens, you think they’d degrade their model quality like that?

Reasoning messages would be lost immediately after a single tool call, unless you mean they sometimes go back and strip the reasoning channel retroactively, but that would increase costs via cache invalidation. I just don’t see any way it would make sense for them to do.

And wouldn’t this be noticeable by reasoning tokens not being accounted for in the context window usage?