SessionModelCostAPI · wallCodeContext replay

agent-rsTeo · Claude Code · 9 subagentsfable-5$5.187m 13s10m 31s wall+164−6494%

ai-frontendOmer · Claude Code · 4 subagentsclaude-opus-4-8$31.7941m 02s1h 26m wall+412−18895%

pulse-webCorey · Cursorgpt-5.5$2.844m 51s22m 09s wall+96−3191%

risk-evalsRamil · Codex · 15 subagentsgemini-3-flash-preview$0.922m 18s3m 02s wall+0−085%

chaos-coreMichael · Claude Code · 2 subagentsclaude-opus-4-8$12.4018m 44s51m 12s wall+230−41293%

capture-proxyYonatan · Geminigemini-3-flash-preview$1.163m 40s9m 55s wall+58−1278%

Cache read is the quiet hero: ~93% of context tokens are replayed, already-paid-for context. The leak is sessions that rebuild it fresh.

Corey hits the 5h window ceiling 3:40 PM todayat current burn · forecast, not fate — a model switch resets it

Omer hits the 5h window ceiling 6:05 PM todayat current burn · forecast, not fate — a model switch resets it

Teo hits the Opus weekly ceiling Thuat current burn · forecast, not fate — a model switch resets it

PersonPlan5h windowWeekOpus · wkResetsBurn · 7D

Omer GoldbergMax 20×82%64%71%1h 12m

Teo DeleanuMax 20×44%71%88%3h 40m

Ramil AmirovAPI——$212 this weekpay-as-you-go

Orr BibringMax 5×61%39%28%0h 48m

Michael KublinMax 5×18%33%12%4h 05m

Corey RicePro93%78%—0h 22m

Yonatan HaimovitzPro37%52%—2h 58m

Shahzad MughalPro9%14%—4h 51m

Mirrors what each person sees in /usage— quota windows, reset timers, burn — plus the forecast the terminal can’t make: who hits a ceiling, and when.