0.148.0 sends prompt_cache_retention on gpt-5.6-sol and every turn fails

Resolved 💬 42 comments Opened Aug 19, 2026 by samkarazi Closed Aug 19, 2026

💡 Likely answer: A maintainer (github-actions[bot], contributor) responded on this thread — see the highlighted reply below.

I updated to 0.148.0 today and every single turn now fails with the below.

{
  "error": {
    "message": "prompt_cache_retention is not supported on this model",
    "type": "invalid_request_error",
    "param": "prompt_cache_retention",
    "code": "invalid_parameter"
  }
}

Codex is still sending prompt_cache_retention on gpt-5.6-sol. That model does not accept it anymore.

I checked my own rollout logs. Same machine, same model, same config, same auth. The only thing that changed is the version.

0.147.0 + gpt-5.6-sol = 10 sessions, 45 turns, 0 errors (Aug 14-18)
0.148.0 + gpt-5.6-sol = 1 session, 3 turns, 3 errors (Aug 19)

On the failing day my update check finished at 02:12Z, the first 0.148.0 session started at 02:14Z, and all 3 of its turns errored at 02:16Z, 02:34Z and 02:41Z.

To reproduce:

1- npm install -g @openai/codex@0.148.0
2- set model = "gpt-5.6-sol" in config.toml
3- send any prompt

Environment: Codex CLI 0.148.0 from npm, win32-x64, Windows 11 10.0.26200, Node v22.22.3, model_reasoning_effort xhigh, ChatGPT account auth and not an API key. Default ChatGPT backend, I did not test Azure or a custom model_provider.

Just to be clear this is NOT the 0.120.0 compaction bug (#17809 to #17819, all closed). Those said "Unknown parameter" and only happened during compaction. Mine says "is not supported on this model" and it happens on every normal turn. My first failure was at 154k tokens in a 258k window so there was no compaction involved.

One thing that makes this confusing to diagnose. When a turn has already started tool calls the work still finishes, but the final message never comes back, so it reads as stuck instead of failed. A turn with no tool calls dies in 4 seconds.

The fix is to gate prompt_cache_retention on the model instead of always sending it. Your own repo already says this in codex-rs/skills/src/assets/samples/openai-docs/references/upgrading-to-gpt-5p6-sol.md:

Treat the older prompt_cache_retention shape as deprecated

Same pattern as #31882 where use_responses_lite and multi_agent_version are hardcoded for sol/terra/luna. A shared capability table would close both.

Workaround for anyone else hitting this, downgrade with npm install -g @openai/codex@0.147.0. Switching to gpt-5.5 also works because that model still accepts the parameter.

0.149.0-alpha.1 was cut at 2026-08-19T00:54Z and I do not see a related commit on main, so I expect it is still broken.

Related: #31882, #25604, #35300.

Thank you

View original on GitHub ↗

42 Comments

github-actions[bot] contributor · 9 days ago

Potential duplicates detected. Please review them and close your issue if it is a duplicate.

#39392

Powered by Codex Action

Water-Run · 9 days ago

Same Issue

leejaen · 9 days ago

Same issue here—it's blocking my workflow. Unfortunately, I haven't found a workaround yet.

igwen6w · 9 days ago

Same Issue

winspyminc · 9 days ago

Same issue here!!!

jieyangxchen · 9 days ago

Same Issue

jasonxiang1990666-web · 9 days ago

same issue !!!! pls fix asap

Water-Run · 9 days ago

Same Issue

OpenAI, don't you test before releasing? Such a basic error is truly baffling.
Given your GPT-5.6-Sol's excessive safety alignment and over-cautiousness, it should have been impossible for it to pass its own readiness check for release.

Xbclid · 9 days ago

lol same issue

AdrianZhang1011 · 9 days ago

same issue, even i changed model from 5.6 sol to 5.5

W313997791 · 9 days ago

same issue,i tried to open another chat window,but it does not disappear

J-H99 · 9 days ago

Same Issue!

Simpidbit · 9 days ago

same issue!

jiayijiayou · 9 days ago

same issue!

Nulnull · 9 days ago

same issue!!!

linyuiu · 9 days ago

same issue!!!

SAIL-Fang · 9 days ago

Same Issue!
It seems that this issue only occurs when multiple windows are running simultaneously. I have three sessions, one of which is stable, while the other two keep encountering this error.

nkzhengwt · 9 days ago

Same Issue!

Mikivishy · 9 days ago

same issue

Darkfeast-x · 9 days ago

payload:
filter:

models:
name: "gpt-5.6-*"

protocol: "codex"
from-protocol: "responses"
params:

"prompt_cache_retention"

可以在cpa中先临时把这个参数过滤掉等后续修复

lirc571 · 9 days ago

Removed prompt_cache_retention from client config but still bumping into the same error frequently

Jimmy-0541 · 9 days ago

Same Issue

ryanwhere · 9 days ago

same issue

xMarKun · 9 days ago

same issue

MikkoAyaka · 9 days ago

same issue

SamSaffron · 9 days ago

yikes, my workaroud here is: https://github.com/SamSaffron/term-llm/commit/26b16336d5db8b1bf6004dee949ec8df3bb72f56

so if you need something urgently a 10 line patch to codex would probably do the same thing while oai sort out the issue.

steinsgo · 9 days ago

same issue

ColdCodeHah · 9 days ago

Same issue,Every conversation is 100% reproduced.

SamSaffron · 9 days ago

For the record I don't think anything is being sent, this is just some weird internal routing thing. the instruction is being injected server side.

whatevertuzi · 9 days ago

same issue

PhyCoe · 9 days ago

codex-cli 0.147.0-alpha.6.6 version
gpt-5.6-sol
same issue

andone-07 · 9 days ago

Same issue!

bxwhite · 9 days ago

Same issue

Water-Run · 9 days ago

codex-cli 0.147.0-alpha.6.6 version gpt-5.6-sol same issue

v0.147.0 also reproduces this issue, too. But it's intermittent.

jjeong-lee · 9 days ago

gpt-5.6-terra, luna same issue!!

winspyminc · 9 days ago

I re-installed the previous version codex but still got the error code. All down!!!

yichenwang100 · 9 days ago

same issue on the pro plan. Could not complete even one session. while "https://status.openai.com/" says it is fully operational.

everyonelovealex · 9 days ago

gpt-5.6-sol,gpt-5.5, same issue!!

Alooswr · 9 days ago

Additional sanitized data point from Windows with gpt-5.6-luna:

Environment

Codex CLI: 0.148.0 from npm
Platform: Windows x64
Authentication/provider: ChatGPT account auth, official default Codex backend
Model: gpt-5.6-luna
Reasoning effort: high
Local config.toml contains neither prompt_cache_retention nor prompt_cache_options

Failure pattern

During one agent turn, the official endpoint returned eight successful HTTP 200 continuation responses between approximately 2026-08-19 06:05:28Z and 06:07:20Z, followed by HTTP 400 at 06:07:24Z:

{
  "error": {
    "message": "prompt_cache_retention is not supported on this model",
    "type": "invalid_request_error",
    "param": "prompt_cache_retention",
    "code": "invalid_parameter"
  }
}

This makes the failure intermittent within the same tool-using turn rather than a deterministic startup/configuration failure.

Local request-shape control

I also redirected the same 0.148.0 binary to a loopback-only Responses endpoint using an ephemeral run (--ephemeral --ignore-user-config --ignore-rules) and captured only the top-level JSON field names. For gpt-5.6-luna, Codex generated:

client_metadata
include
input
model
parallel_tool_calls
prompt_cache_key
reasoning
store
stream
text
tool_choice

Relevant values/presence:

model: "gpt-5.6-luna"
prompt_cache_key: present
prompt_cache_retention: absent
prompt_cache_options: absent
store: false
reasoning.context: "all_turns"

The actual failure was observed against https://chatgpt.com/backend-api/codex/responses, not the loopback provider. This control strongly supports the theory that the deprecated retention field is being injected or transformed in the ChatGPT/Codex backend routing path (or another provider-specific layer), rather than coming from the user's config or the provider-agnostic request body.

The current GPT-5.6 migration guidance says to replace prompt_cache_retention with prompt_cache_options.ttl:
https://developers.openai.com/api/docs/guides/latest-model

No prompts, auth headers, account identifiers, thread IDs, or full logs are included here.

Simpidbit · 9 days ago

https://github.com/Simpidbit/retrycodex/
Just retry, this project works

etraut-openai contributor · 8 days ago

This should be resolved now. It was a server-side configuration issue.

yuliyantsvetkov · 7 days ago

Still occurs.