Codex Pulse
What Codex users are asking right now
The most reported Codex usage problem this week is “model at capacity” errors, with 38 public GitHub reports in the last 7 days. Token accounting & compaction is climbing: 5 reports, up from 0 the week before.
Latest questions by topic
- “Model at capacity” errors38 this week · steady[Bug][macOS][Work] Deterministic reasoning regression across high-end models + repeated capacity errors; Chat mode remains correct
- Unexpected usage depletion27 this week · steadyRequest for an investigation into Codex GPT-6 task quality and high usage, and for quota compensation
- Usage display & /status16 this week · steadyCodex App MacOS /status says Rate limit: Unavailable while Settings shows usage
- Plan changes & allowance9 this week · ↑ risingUpgraded to Pro hit limit after running one command!!!
- Requests to change limits7 this week · steadyFeature request: remove the 5-hour usage limit
- Hitting the limit5 this week · steadyapp-server: turn ends with `error` then `turn/completed` when the account usage limit is reached, so drivers relaunch in a tight loop
- Token accounting & compaction5 this week · ↑ risingHosted search usage causes premature auto-compaction
- Banked & instant resets5 this week · ↑ rising[Pro 20x][macOS][GPT-6 Astra Ultra] One ~8-hour task exhausted two usage-reset credits
This week at a glance
- Most reported
- “Model at capacity” errors
- 38 reports in 7 days
- Fastest rising
- Token accounting & compaction
- 0 → 5 week over week
- New recurring terms
- None this week
- No unfamiliar name recurred across 3+ reports
What changed: 131 usage-related issues were filed in the last 7 days, against 174 the week before. Overall volume is roughly steady. 17 came in over the last 24 hours.
Topics this week
Ordered by reports in the last 7 days. Open a topic for the actual questions, links to each GitHub issue, and what Codex Usage already covers.
“Model at capacity” errorsSteady38 this week · 50 the week before
38 reports in the last 7 days, 50 the week before. Roughly steady. 2 in the last 24 hours.
Recent questions
- #47794[Bug][macOS][Work] Deterministic reasoning regression across high-end models + repeated capacity errors; Chat mode remains correct
- #47790Selected model is at capacity. Please try a different model.
- #47304Repeated "Selected model is at capacity" / server_overloaded across models despite available Pro usage
- #47238[Bug] Pro $200 account always shows "Selected model is at capacity" despite 69% weekly limit remaining
Models or modes named in these titles: Astra (4), Reasoning effort (3), Sol (3), Luna (3).
What this shows
About as many reports as the week before. This is a steady, recurring question, not a new spike.
Evidence type: community-reported GitHub issues. Not official, not measured by Codex Usage.
Unexpected usage depletionSteady27 this week · 32 the week before
27 reports in the last 7 days, 32 the week before. Roughly steady. 3 in the last 24 hours.
Recent questions
- #47880Request for an investigation into Codex GPT-6 task quality and high usage, and for quota compensation
- #47878[CLI][Astra] Research loop consumes excessive quota and incorrectly re-applies an old pause instruction
- #47823many users commenting on excessive usage
- #4763798% of fresh 5-hour Codex usage consumed within minutes — possible metering issue
Models or modes named in these titles: Astra (11), Sol (5), Reasoning effort (4), Luna (4).
What this shows
About as many reports as the week before. This is a steady, recurring question, not a new spike.
Evidence type: community-reported GitHub issues. Not official, not measured by Codex Usage.
Usage display & /statusSteady16 this week · 24 the week before
16 reports in the last 7 days, 24 the week before. Roughly steady. 6 in the last 24 hours.
Recent questions
Models or modes named in these titles: Spark (8).
What this shows
About as many reports as the week before. This is a steady, recurring question, not a new spike.
Evidence type: community-reported GitHub issues. Not official, not measured by Codex Usage.
Plan changes & allowanceRising9 this week · 5 the week before
9 reports in the last 7 days, up from 5 the week before (+80%). 2 in the last 24 hours.
Recent questions
- #47893Upgraded to Pro hit limit after running one command!!!
- #47869VS Code: Work locally selector does not open; subscription endpoints return 403
- #47586[Bug] Free weekly quota carries over after upgrading to Plus, leaving paid Codex capacity at 0%
- #47516Codex: silent downgrade to gpt-4o - selected model never actually served (3 models tested, 8/8 valid answers)
What this shows
Reports on this topic are showing up more often in the public GitHub issue stream than the week before. That doesn't establish a platform-wide change.
Evidence type: community-reported GitHub issues. Not official, not measured by Codex Usage.
Requests to change limitsSteady7 this week · 7 the week before
7 reports in the last 7 days, 7 the week before. Roughly steady. 2 in the last 24 hours.
Recent questions
- #47928Feature request: remove the 5-hour usage limit
- #47740Add a setting to disable CLI usage-limit warnings
- #47500The "5h limit" should be canceled during the last 5 hours of the "Weekly limit" for Plus subscribers.
- #47261Allow continued Codex usage after the 5-hour limit by consuming weekly quota at an increased rate
Models or modes named in these titles: Sol (1).
What this shows
About as many reports as the week before. This is a steady, recurring question, not a new spike.
Evidence type: community-reported GitHub issues. Not official, not measured by Codex Usage.
Hitting the limitSteady5 this week · 8 the week before
5 reports in the last 7 days, 8 the week before. Roughly steady. 1 in the last 24 hours.
Recent questions
- #47728app-server: turn ends with `error` then `turn/completed` when the account usage limit is reached, so drivers relaunch in a tight loop
- #47503Chat Bar (macOS) incorrectly blocked by “Codex/Work usage limit” while regular Chat works (Go plan)
- #47356Rate limit error on Pro
- #47309MultiAgentV2 usage hints can exceed the 10K model-visible item limit
Models or modes named in these titles: Luna (1).
What this shows
About as many reports as the week before. This is a steady, recurring question, not a new spike.
Evidence type: community-reported GitHub issues. Not official, not measured by Codex Usage.
Token accounting & compactionRising5 this week · 0 the week before
5 reports in the last 7 days, up from 0 the week before. 1 in the last 24 hours.
Recent questions
- #47825Hosted search usage causes premature auto-compaction
- #47003Auto-compaction request usage is never added to total_token_usage in rollouts (not recorded at all in 0.147)
- #46975WebSocket startup prewarm reports ~9k uncached input tokens omitted from turn usage
- #46948[macOS Desktop 26.915.31945] Automatic thread_description used 267,264 input tokens, then failed with invalid_encrypted_content
What this shows
Reports on this topic are showing up more often in the public GitHub issue stream than the week before. That doesn't establish a platform-wide change.
Evidence type: community-reported GitHub issues. Not official, not measured by Codex Usage.
Banked & instant resetsRising5 this week · 1 the week before
5 reports in the last 7 days, up from 1 the week before. None in the last 24 hours.
Recent questions
Models or modes named in these titles: Astra (2), Reasoning effort (1), Spark (1).
What this shows
Reports on this topic are showing up more often in the public GitHub issue stream than the week before. That doesn't establish a platform-wide change.
Evidence type: community-reported GitHub issues. Not official, not measured by Codex Usage.
Reset timing & windowsSteady3 this week · 5 the week before
3 reports in the last 7 days, 5 the week before. Roughly steady. None in the last 24 hours.
Recent questions
What this shows
About as many reports as the week before. This is a steady, recurring question, not a new spike.
Evidence type: community-reported GitHub issues. Not official, not measured by Codex Usage.
Luna ReserveFew reports1 this week · 3 the week before
1 report this week, 3 the week before. Too few to call a trend. None in the last 24 hours.
Recent questions
- #46421Unable to switch out of the exhausted Luna Reserve mode in an old resumed session
- #46227Luna Reserve rollout is missing from What's New / changelog, so eligible Pro users discover it only after quota exhaustion
- #46181Luna reserve not working in CLI or desktop app
- #45132app-server clients cannot use Luna Reserve: supportsLunaReserve exists but has no accept/redeem action
Models or modes named in these titles: Luna (4).
What this shows
Only a handful of reports. Treat these as individual accounts, not a pattern.
Evidence type: community-reported GitHub issues. Not official, not measured by Codex Usage.
How the Pulse works, and what it can't tell you
- Source: public issues in the openai/codex repository. Everything OpenAI labels
rate-limits, plus issues whose titles clearly mention usage, quota, limits or resets. Titles about CPU usage, context limits or usage policy are left out. - Windows: the last 24 hours, the last 7 days, and the 7 days before that, counted back from the update time. Of 327 issues checked in that span, 305 were usage-related.
- Topics: each issue title is matched against fixed rules, with no AI in the loop. An issue lands in one topic only. 54 usage-related issues didn't fit any topic and aren't shown in one.
- Emerging terms: a name that shows up in at least three different issue titles, on more than one day, and isn't part of what this site already covers.
- Trends: a topic is only called rising or cooling when it moves by at least 3 reports and at least half of the previous week. If neither week reaches 5 reports, there's no trend call, and percentages only appear once the previous week has at least 5.
- Limits: GitHub issues come from people motivated enough to file one. The counts show what's being reported, not how many Codex users are affected, and they don't confirm any change on OpenAI's side. Everything on this page is community-reported evidence.
- Freshness: the snapshot refreshes automatically about every 4 hours. The same data is available as JSON.