OpenAI Admits Codex Cache Hit Rate Dropped After Users Report Quota Anomalies, Promises Compensation

On Aug 22, OpenAI Codex users reported a surge of abnormal quota consumption. Product lead Tibo Inoue's initial denial was rebutted by community evidence, and the incident closed for now with OpenAI acknowledging a drop in cache hit rates and promising compensation.

Confirmed

• MaziyarPanahi reported burning 60% of his weekly quota in a single day with token usage identical to normal; Codex team members he tagged said they found no issue. RileyRalmuto, who has long built the same app with Codex, measured at least 20% higher consumption for the same workload, suspecting either a more token-hungry model or a quiet quota cut.
• Inoue later partially acknowledged the problem, saying cache hit rates for some users this week fell below previously stable levels, accelerating consumption; the team is investigating with an update expected the next day, and compensation will be offered.
• Inoue's initial denial, which included accusing users of fraud, was rebutted by community evidence and flagged by X's Community Notes.

Unconfirmed

• The root cause (model behavior vs. caching infrastructure) awaits the official investigation.
• RileyRalmuto felt consumption slightly improved afterward but could not confirm a fix.
• The scope and form of compensation remain unclear.

Why it matters

• The episode shows how X's Community Notes can instantly correct vendor claims; executive denial holds little ground against user evidence.
• For developers relying on Codex, quota consumption directly affects cost and availability—fluctuations in infrastructure parameters like cache hit rates materially hurt users, and transparent communication beats denial.

2026-08-22 ~ 2026-08-23 · 7 related posts

Full story(10 episodes)→

• Episode 1: OpenAI Codex Active Users Exceed 15 Million with Reset Quotas(2026-08-12, 4 posts)
• Episode 2: OpenAI Codex Users Report Quotas Drained Rapidly, Quality Complaints Mount(2026-08-19, 4 posts)
• Episode 3: OpenAI Codex Hits 20 Million Active Users, Resets Usage Credits(2026-08-21, 4 posts)
• Episode 4: OpenAI Admits Codex Cache Hit Rate Dropped After Users Report Quota Anomalies, Promises Compensation(2026-08-22, 7 posts)
• Episode 5: OpenAI Users Complain of Frequent Usage Limits Without Reset(2026-08-22, 2 posts)
• Episode 6: OpenAI Admits Three Codex Bugs, Resets All Paid Usage(2026-08-23, 6 posts)
• Episode 7: OpenAI Restores 5-Hour Codex and Work Usage Limits for ChatGPT Plus(2026-08-25, 4 posts)
• Episode 8: OpenAI Codex Users Report Quotas Draining Abnormally Fast(2026-09-05, 5 posts)
• Episode 9: Codex Users Report Runaway Quota Burn, Blame New Astra Model(2026-09-16, 8 posts)
• Episode 10: Users Flee $200 Codex Plan as Credits Burn in Days, Claiming Claude Offers 5x Value(2026-09-27, 3 posts)

• [source] User Reports Abnormal Usage Drain in OpenAI Codex — MaziyarPanahi · 2026-08-22
• OpenAI exec denies Codex throttling, gets community-noted with evidence — gaganghotra_ · 2026-08-22
• [source] Users Question Faster Token Burn in ChatGPT/Codex Despite Official Denials — RileyRalmuto · 2026-08-22
• Dev reports ChatGPT/Codex burning through usage allowance 20% faster — RileyRalmuto · 2026-08-22
• [source] OpenAI Codex Cache Hit Rate Drops, Causing Faster Usage Drain — ChrisGPT · 2026-08-22
• User Reports Codex Usage Drain: Loses 60% of Weekly Quota in One Day — MaziyarPanahi · 2026-08-22
• User reports Codex usage drain bug: 60% of weekly quota gone in a day — MaziyarPanahi · 2026-08-23