The /cost command can accurately calculate Token consumption, but the statusline data is all 0

[zhufeizzz]

Claude Version：2.1.23

Claude Project JSONL File
{"input_tokens":0,"output_tokens":0}

{"parentUuid":"9e535cee-7627-4fc2-a6c0-710dbf737fd0","isSidechain":true,"userType":"external","cwd":"E:\\Demo穿刺\\ClaudeCodeUntrust","sessionId":"9671ff51-a277-46a2-854c-c6df041557ab","version":"2.1.23","gitBranch":"","agentId":"aprompt_suggestion-2f66c5","message":{"id":"msg_2026012914205274f33cbf2ede4cc0","type":"message","role":"assistant","model":"glm-4.7","content":[{"type":"text","text":"你好！很高兴见到你。我可以帮助你完成软件开发任务，比如编写代码、调试问题、创建架构文档等。有什么我可以帮助你的吗？"}],"stop_reason":null,"stop_sequence":null,"usage":{"input_tokens":0,"output_tokens":0}},"type":"assistant","uuid":"282a2679-84d3-4402-b7ec-6090e2e97267","timestamp":"2026-01-29T06:20:57.664Z"}

[PhantomLasagne]

cross-posted from another issue thread

I'm having similar issues, though, strangely, the context will display correctly anywhere from 20-50% of the time. In my statusline I'm using context length, tokens cached (always displays accurately) and context % (usable). I've attempted a few fixes with changing settings and saving settings with/without CC open with no improvement.

Workspace:

• VS Code (Win11 install run through WSL2)
• Claude Code CLI
• GLM-4.7 / Minimax-M2.1

Incorrect statusline display:

Model: glm-4.7 | v2.1.33 | Style: Explanatory |
⎇ main | 𖠰 main | (+93,-390) |
Session: 7m | Total: 428.7k | In: 70.8k Out: 3.6k |
Ctx: 0 | Cached: 354.3k | Ctx(u): 100.0% |

And a perfect example, while typing this the statusline started to display correctly:

Model: glm-4.7 | v2.1.33 | Style: Explanatory |
⎇ main | 𖠰 main | (+154,-441) |
Session: 16m | Total: 1.0M | In: 143.9k Out: 6.0k |
Ctx: 87.0k | Cached: 874.5k | Ctx(u): 45.6% |