Claude 요금제별 사용량 한도

Hammered the $100 Codex plan all month with parallel agents and deep coding sessions. This is the first time I've hit below 50%. The limits are clearly designed for builders who ship real work, not just flirt with the model. Been reverse-engineering usage patterns for my SaaS tools and it's wild how much variance there is between documented limits and actual throttling. Codex feels more aggressive than Claude but haven't seen concrete numbers anywhere either. Seat pricing matters for power users. For service-biz operators the number that matters is per-call cost. A receptionist agent at $200/mo on Max 20x is invisible. Same agent at API rates after the cap blows the unit economics. I observed $35 in 5.5 token spend (~35M tokens, ~90% cache) yesterday hit the weekly limit by 2% ($100/mo plan). I extrapolated that to ~$6500 in potential max value. But, limits on the $100 plan are doubled until the end of may, so long term more like ~$3000. Those numbers are outdated because the Claude plan limits have been increased after the release of Opus 4.7 due to the new tokenizer. I think the 20x plan is closer to $3600/mo now. Guys don't forget, the it's a combination of input and output. And they make 50% profit on API cost. So the real subsidy is about 4-7x if all users, used all of their capacity every single day, every single week, month. This also isn't the full story: free cache hits on the plans is a massive cost saver on top of pure token price subsidy. In practice that's where I get dinged hardest on API vs subscription. We should all prepare for the day when subscriptions are gone, and we're paying per token. These companies cannot continue to operate like this... losing money on every subscription. reverse engineering plan limits is fun but the numbers shift all the time. anthropic and openai both tune them per region, time of day, and load. screenshot today does not match next week. nice exercise, weak signal