I’m trying to understand whether this is a known GPT-5.5 / Responses API issue, an account/project/routing issue, or something unusual about my request shape. I run a long-context agent harness from New Zealand, usually routed through Auckland / ANZ Cloudflare paths. The workload is repeated same-c…

I’ve made good progress so far. The first smaller win was identifying a previously unknown source of body mutation through my existing fingerprinting mechanisms. Though cleaning that up was just an improvement to the telemetry and the ability to diagnose this further itself. That is to say, my app…

So far I’ve made significant modifications to the way my application handles the context windows. It seems the invariant is that whenever a cache write occurs, it is for the full context window, including and right up to the very last byte of the tail content, no matter the volatility, and it refus…

[image] This is what a day of light usage looks like now that I’ve fixed most of the body churn, but I’m still desperately fighting the… full fingerprinting issues that I’m encountering. Every time I get an unscheduled cache bust, I have to then attempt to create a clean cache-write event that I …

Changes in the API giving symptoms you didn’t have before mean that OpenAI should investigate and solve the problem. The contract and the pricing are that the cache system shall work as described: matching any previous context length, up to the point of divergence. When I say the “bulk” - I mean th…

[image] Yeah, I think it’s pretty clear that the expectation and contract of engaging their services has not been honoured here. I mean, my architecture didn’t change significantly. This is what the distribution looked like before the whatever changed at the infrastructure level. And this is wher…

Given the dollar figures you are talking about here: something you should not need to do but may provide relief: USA-located virtual private server, acting as a proxy relay for OpenAI API calls. And not a Cloudflare worker, which they populate around the world. When that magically starts working,…

That’s actually similar to how I solved the cloudflare 520 issues. I tunneled into a German VPS and have been proxying my traffic from there, and it’s been smooth sailing ever since. But that did absolutely nothing to fix the cache. The thing I’m still scratching my head about is why I’m getting th…

This is OpenAI messing up, also happening to me. I’ve been overcharged hundreds of bucks in recent days where cached_tokens:0 when they should be cached_tokens:180.000. I’m raising an issue with support@openai.com . Note that this was a recurring issue with gpt-5.3 as well. They just don’t bother …

Yeah, this is just absurd. They are not delivering the service that I agreed to when I signed up. I was told that I’d be getting prompt cache, and they silently changed or broke that, and now I’m hemorrhaging money. I’ve spent ten times as much diagnosing and redesigning my project architecture to …

Still radio silence from their support team as well. Supposedly it’s been escalated multiple times, but beyond some “very reassuring prose” from their chatbot support, absolutely nothing.

Persistent 0% prompt cache hits on GPT-5.5 with Auckland NZ Cloudflare 520s complicating every workaround

API

slackermanz June 25, 2026, 3:46am 26

Just figured I’d record this one here as part of the cluster.

Topic		Replies	Views
Prompt cache: documented byte-prefix matching does not occur on gpt-5.4 / gpt-5.5 when trailing user content exceeds ~500 tokens Bugs gpt-5-5	6	251	June 27, 2026
We need to talk about prompt caching Feedback prompt-caching , responses-api , chat-completions-api	1	923	October 25, 2025
Perceived Drop in GPT-5 Quality Over the Last Few Weeks Codex gpt-5-codex , gpt-5-5	32	1814	June 3, 2026
Cache not caching more than 1024 tokens (expected: increments of 128 tokens) Bugs prompt-caching	6	468	November 14, 2024
Structured Outputs not reliable with GPT-4o-mini and GPT-4o API structured-output	38	9819	January 23, 2025

Persistent 0% prompt cache hits on GPT-5.5 with Auckland NZ Cloudflare 520s complicating every workaround

Related topics