Distributed Codex Radar is live. Everyone can contribute a run so model degradation has nowhere to hide. The monthly ladder's top three receive exclusive custom 3D-printed radar trophies.
Quick links
Some promotional reset cards may expire soon. Open Settings → Usage left → Available resets, then expand Usage limit reset to see the expiry time. Use cards before they expire.
Open Codex Settings → Configuration → Model features → Available reasoning efforts, then select Max. It will appear in model controls for supported models.
Register, then request credits from Resources by Jul 17, 12:00 PM PT. Limited supply and approval apply. Projects are due Jul 21, 5:00 PM PT; promotional credits must be used by Jul 31.
Tibo's 9M action directly restored weekly usage to 100%; it did not grant a new redeemable banked reset. Some users are now using existing banked resets or checking their expiry, but his latest posts and replies add no official promise of another card grant.
At 12:14 CST Tibo announced another reset for Codex and ChatGPT Work users and said the weekly usage limit should return to 100% within minutes. Multiple users then reported restored quota. Fresh reports include a resumed 27M-token session consuming another 2%, xhigh dropping about 1% per minute, and one 20-hour run exhausting a weekly allowance while producing an unwanted rewrite. These remain scattered anecdotes; no new reset promise followed, and OpenAI Status reports all systems operational.
The 5h limit is temporarily paused, so Quota Radar currently shows only the active 7d quota. GPT-5.6 estimates include an estimated cache-write surcharge.
Using the latest valid 7d calibration (2026-07-16-am) from a 20x Pro US account. The 5h limit is temporarily paused and is not displayed. Formula: 20x Pro 7d = $214.26 / 15% * 100 = $1,428.41; the API-equivalent cost includes the estimated cache-write surcharge.
Method: Sol, Terra, and Luna use low reasoning effort with the same count-from-1-to-1024 output task. Standard and Fast each run three independent samples and use the arithmetic mean; order alternates, up to three model pairs run concurrently, and each pair still runs Standard/Fast serially. Every sample uses a fresh app-server and ephemeral thread plus a random nonce. Complete outputs with cached_input_tokens = 0 are preferred. For this Jul 14 run, slow samples and one complete Terra cache-mismatched pair are retained as real same-day experience; incomplete outputs are rejected and rerun. TTFT is delay to the first visible token, TPS is sustained token generation speed, and E2E is total time from submission to the complete result. Headline acceleration values are calculated as mean Standard divided by mean Fast; per-model E2E acceleration uses that model's Standard divided by Fast.
Scoring: for each task, use only its latest publishable valid result, then weight tasks equally to calculate pass rate and IQ; repeated runs of one task are not averaged. Distributed Radar uses a live community leaderboard methodology, so its scores may differ. Coverage varies by model; cost and time are per-task averages. See Distributed Radar for broader coverage
Overall IQ uses every publishable task available for each model. This original shared 10-task set remains only for row-by-row comparison.
ytt-jsonpath-query-api
abs-module-cache-flags
httpx-multipart-response-parsing
bandit-incremental-cache-control
ipython-session-bundle-replay
ofetch-per-origin-circuit-breaker
obsidian-linter-link-format-conversion
csstree-shorthand-expansion-compression
fd-deterministic-multi-key-sorting
oxvg-structural-selector-preservation
Archived previous-generation curves and daily comparison table. This section is no longer updated.
WeChat Radar Group
WeChat public account
JSON is current state; RSS focuses on official windows and reset alerts.