Context percentage incorrect for Opus with 1M token context window

[PaulRBerg]

Problem

Since Claude Code v2.1.75, the default model shipped with the Max subscription is Opus 4.6 with a 1M token context window. ccstatusline doesn't calculate the context usage percentage correctly for this configuration — it likely still assumes the previous context window size, resulting in inaccurate percentage values in the status line.

Note

The exact model ID is claude-opus-4-6[1m]. You can verify this by checking claude --version or looking at the model info in the system prompt of any active session.

[sirmalloc]

@PaulRBerg Can you add a custom command widget, then edit the command and set it to cat > status.json, save your config, run claude, and send me the file it creates? The context window length is supposed to come in to the statusline from that json, not be inferred from the model name. Seeing what your statusline is receiving will assist in debugging.

[PaulRBerg]

{
  "session_id": "7ee4f3f3-6530-4aae-aaa1-9c89c188ee06",
  "transcript_path": "/Users/prb/.claude/projects/-Users-prb-projects-markdrop/7ee4f3f3-6530-4aae-aaa1-9c89c188ee06.jsonl",
  "cwd": "/Users/prb/projects/markdrop",
  "model": {
    "id": "claude-opus-4-6[1m]",
    "display_name": "Opus 4.6 (1M context)"
  },
  "workspace": {
    "current_dir": "/Users/prb/projects/markdrop",
    "project_dir": "/Users/prb/projects/markdrop"
  },
  "version": "2.1.76",
  "output_style": {
    "name": "default"
  },
  "cost": {
    "total_cost_usd": 0,
    "total_duration_ms": 936,
    "total_api_duration_ms": 0,
    "total_lines_added": 0,
    "total_lines_removed": 0
  },
  "context_window": {
    "total_input_tokens": 0,
    "total_output_tokens": 0,
    "context_window_size": 1000000,
    "current_usage": null,
    "used_percentage": null,
    "remaining_percentage": null
  },
  "exceeds_200k_tokens": false,
  "vim": {
    "mode": "INSERT"
  }
}

[sirmalloc]

@PaulRBerg Thanks. I'm not able to reproduce what you're seeing. Is it possible you could send me a screenshot of the statusline showing the incorrect values, and send a copy of the status.json file that was output during that screenshot?

For the Context % widget, the displayed value comes from this chain:

• First choice: status JSON -> context_window.used_percentage. That is used directly in [ContextPercentage.ts:47]
• If used_percentage is missing, ccstatusline tries to derive it from context_window.current_usage and context_window.context_window_size in [context-window.ts:27]
• If the status payload still doesn’t provide usable usage info, it falls back to transcript metrics: tokenMetrics.contextLength / maxTokens in [ContextPercentage.ts:52] and the shared helper [context-percentage.ts:12]

That fallback contextLength comes from the most recent main-chain transcript entry in [jsonl-metrics.ts:151], specifically input_tokens + cache_read_input_tokens + cache_creation_input_tokens at [jsonl-metrics.ts:189]. The denominator maxTokens comes from context_window.context_window_size if present, otherwise inferred from model id/display name, otherwise defaults to 200k in [model-context.ts:76]

Context % (usable) uses the same source for usage, but divides by usableTokens = 80% of max in [ContextPercentageUsable.ts:49]

My gut feeling says that Anthropic is sending us an incorrect used_percentage.

[PaulRBerg]

thanks for the quick replies @sirmalloc and for your assistance!

Toggle to see `status.json`

{"session_id":"a0c52ee6-747d-4bc4-a362-adb8c6ce26cd","transcript_path":"/Users/prb/.claude/projects/-Users-prb-projects-markdrop/a0c52ee6-747d-4bc4-a362-adb8c6ce26cd.jsonl","cwd":"/Users/prb/projects/markdrop","model":{"id":"claude-opus-4-6[1m]","display_name":"Opus 4.6 (1M context)"},"workspace":{"current_dir":"/Users/prb/projects/markdrop","project_dir":"/Users/prb/projects/markdrop"},"version":"2.1.76","output_style":{"name":"default"},"cost":{"total_cost_usd":0.17794925,"total_duration_ms":32958,"total_api_duration_ms":4795,"total_lines_added":0,"total_lines_removed":0},"context_window":{"total_input_tokens":6,"total_output_tokens":21,"context_window_size":1000000,"current_usage":{"input_tokens":3,"output_tokens":8,"cache_creation_input_tokens":19,"cache_read_input_tokens":33790},"used_percentage":3,"remaining_percentage":97},"exceeds_200k_tokens":false,"vim":{"mode":"INSERT"}}

[sirmalloc]

@PaulRBerg Please double check how you are invoking the statusline in ~/.claude/settings.json to ensure it's launching via either npx or bunx, and it has a @latest on the end. If not and you did a global install it's potentially out of date. The behavior you are seeing is consistent with a version pre-v2.0.26. The latest version is v2.2.4.

[PaulRBerg]

Yeah I was using a globally installed version. I switched to npx -y ccstatusline@latest and it works now.

Sorry for this and appreciate your support 🙏

[FammasMaz]

with a custom api this isnt working, my model name is "provider/claude-opus-4-6" and with the [1m] option added from the models, it still shows 200k context

[sirmalloc]

with a custom api this isnt working, my model name is "provider/claude-opus-4-6" and with the [1m] option added from the models, it still shows 200k context

@FammasMaz Can you explain what you mean by 'custom api'? If the statusline json doesn't have the correct context length, and the model name doesn't have [1m] as part of it, I'm not sure how we would be able to determine the maximum context length.