Claude Code Opus 4.7: 8개 세션에서 16B 캐시 읽기, 포렌식 JSONL 데이터

You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert I have searched existing issues and this hasn't been reported yet This is a single bug report (please file separate reports for different bugs) I am using the latest version of Claude Code What's Wrong? Since March 22 March, 2026, my session limits has been exhausting extremely fast (much faster than before that date), without any real change to my input / usage patterns. What Should Happen? Rate limits on the max 5x plan shouldnt be hit repeatedly under moderate usage Error Messages/Logs For the last **1 hour 58 minutes**, my local Claude Code session logs show: - **Time window (local, Asia/Ho_Chi_Minh):** `2026-03-25 01:00:32` to `2026-03-25 02:58:32` - **Time window (ET / EDT):** `2026-03-24 14:00:32` to `2026-03-24 15:58:32` - **Sessions:** `4` - **API calls:** `53` - **Input tokens:** `85` - **Output tokens:** `69,004` - **Cache creation tokens:** `265,514` - **Cache read tokens:** `5,511,490` - **Total tokens:** `5,846,093` - **Quota-pressure / rate-limit estimate:** `334,603` Formula used from the Python script @hgreene624 : `quota_pressure_estimate = input_tokens + output_tokens + cache_creation_input_tokens` This excludes `cache_read_input_tokens`. For the same **2h 43m** period, my **5x Max plan** usage indicator increased by **11%**. Because this was during a **2x bonus usage window**, that is equivalent to roughly **22% of normal usage**. This was an absurdly small workload for such a high usage limit increase. Steps to Reproduce use 1m context window opus in Claude Code CLI Regular usage which previously would never get beyond 50% of 5 hour limit has now hit 5 hour limit multiple times in the last few days. Weekly limit % is also going up much faster. This appears to be a widespread issue — multiple users across platforms are reporting the same behavior around the same timeframe. Claude Model Opus Is this a regression? Yes, this worked in a previous version Last Working Version No response Claude Code Version 2.1.81 Platform Anthropic API Operating System macOS Terminal/Shell iTerm2 Additional Information Regression in quality of model since Opus 1M context has also been reported and I have experienced this, and I think this may be contributing - repeated backpedalling, duplication of works, errors and retrying, low quality output meaning it has to go back and do it again (auto accept edits mode on) - input token usage is similar, output token usage is enormous. subagents makes it even worse -issue appeared immediately not gradually and does not seem to be isolated to v2.1.81, it seems to be model related. Preflight Checklist What's Wrong? Since March 22 March, 2026, my session limits has been exhausting extremely fast (much faster than before that date), without any real change to my input / usage patterns. What Should Happen? Rate limits on the max 5x plan shouldnt be hit repeatedly under moderate usage Error Messages/Logs Steps to Reproduce use 1m context window opus in Claude Code CLI Regular usage which previously would never get beyond 50% of 5 hour limit has now hit 5 hour limit multiple times in the last few days. Weekly limit % is also going up much faster. This appears to be a widespread issue — multiple users across platforms are reporting the same behavior around the same timeframe. Claude Model Opus Is this a regression? Yes, this worked in a previous version Last Working Version No response Claude Code Version 2.1.81 Platform Anthropic API Operating System macOS Terminal/Shell iTerm2 Additional Information Regression in quality of model since Opus 1M context has also been reported and I have experienced this, and I think this may be contributing - repeated backpedalling, duplication of works, errors and retrying, low quality output meaning it has to go back and do it again (auto accept edits mode on) - input token usage is similar, output token usage is enormous. subagents makes it even worse -issue appeared immediately not gradually and does not seem to be isolated to v2.1.81, it seems to be model related.