I’m trying to determine whether I’ve hit a Codex credit-metering/accounting issue, and I’d particularly like to hear from anyone who has had a similar case successfully escalated.
I have an open OpenAI Support case: #12830680.
The relevant timeline is:
• My final Codex session on 6 August 2026 ended at approximately 17:24 BST, with my purchased credit balance at 0.
• At approximately 17:34 BST, after that session had ended, I purchased 500 additional Codex credits.
• I did not use Codex again that evening or overnight.
• My local .codex session records show no intervening Codex session.
• My next recorded Codex session began at approximately 08:51 BST on 7 August.
• Within seconds of that session starting, Codex’s own local telemetry recorded my purchased credit balance as 29.6829375000 credits.
In other words, approximately 470.317 of the 500 credits purchased the previous evening had already disappeared before my next locally recorded Codex session began.
I have sanitised the relevant token_count records below. I have removed unrelated token counters and project information; the timestamps and credit fields are unchanged.
END OF PREVIOUS CODEX SESSION:
{
“timestamp”:“2026-08-06T16:23:21.459Z”,
“type”:“event_msg”,
“payload”:{
“type”:“token_count”,
“rate_limits”:{
“credits”:{
“has_credits”:false,
“balance”:“0”
},
“plan_type”:“plus”
}
}
}
FIRST CREDIT RECORD IN MY NEXT CODEX SESSION:
{
“timestamp”:“2026-08-07T07:51:39.734Z”,
“type”:“event_msg”,
“payload”:{
“type”:“token_count”,
“rate_limits”:{
“credits”:{
“has_credits”:true,
“balance”:“29.6829375000”
},
“plan_type”:“plus”
}
}
}
The JSONL timestamps are UTC. These correspond to approximately 17:23 BST on 6 August and 08:51 BST on 7 August.
The first balance reading of 29.6829375000 is not an isolated or transient value. It remains at that figure through multiple subsequent token_count events at the beginning of the 7 August session.
The 500-credit purchase was made at approximately 17:34 BST on 6 August, after the earlier session had ended.
This is separate from the question of whether individual Codex tasks can be expensive. I have also examined the token-count events in my local JSONL files and can see how long, context-heavy sessions can consume substantial usage.
What I cannot account for is this overnight gap.
OpenAI Support has acknowledged the timeline, but has repeatedly told me that they cannot access or validate the internal task-level metering records, token usage or telemetry needed to determine what caused the deductions.
Their latest response says they cannot determine whether the missing credits were related to a particular Codex task, another eligible feature using flexible credits, or “another internal usage event”.
They have also declined to restore or refund the disputed credits because they say there is no evidence confirming that the credits were deducted incorrectly.
That leaves me in a difficult position: OpenAI says it cannot verify what consumed the credits, but the inability to verify the deduction is then being used as the reason not to treat it as erroneous.
I am not asking OpenAI to disclose raw internal telemetry to me. I am asking for the account/session records to be reviewed internally by someone capable of determining what actually consumed the balance.
I have also seen other reports of unusually rapid Codex quota/credit depletion, including reports of limits or purchased balances dropping unexpectedly or while users were effectively idle, so I am trying to establish whether this is an isolated account issue or part of a broader metering problem.
Has anyone here had a similar unexplained Codex credit deduction successfully escalated to a team that could review the account/session records?
And if an OpenAI staff member sees this, could Case #12830680 please be routed to the appropriate Codex billing/engineering team for an actual account-level investigation?
I can provide further sanitised extracts from the local JSONL if useful, but I do not want to post the complete files publicly because they contain prompts, code, file paths and other private project information.