Skip to content

GOAT Plan: DeepSeek V4 Flash Cache Read billed ~53% higher than published rate (off-peak usage) #722

Description

@ibpilot

Summary

Description

I've identified a billing discrepancy on the GOAT plan when using DeepSeek V4 Flash. The actual cost for Cache Read tokens is significantly higher than the published rate in the official pricing table, while Input and Output costs match the table exactly.

Expected Behavior

Based on the official GOAT pricing table for DeepSeek V4 Flash:

  • Input: $0.22/M
  • Output: $0.66/M
  • Cache Read: $0.007/M

With my usage (all during off-peak hours):

  • Input: 2.5M × $0.22 = $0.55

  • Output: 0.7874M × $0.66 = $0.5197

  • Cache Read: 83.1M × $0.007 = $0.5817

  • Total expected: $1.65

Actual Behavior

Dashboard shows:

  • DeepSeek V4 Flash: $1.96

  • web_search: $0.01

  • Total actual: $1.97

That's a $0.31 difference

Image Image

Steps to reproduce the issue

  1. Use DeepSeek V4 Flash on the GOAT plan during off-peak hours.
  2. Generate significant Cache Read usage (83.1M tokens over 2 days).
  3. Compare the actual billed amount against the official pricing table.

Supporting Information

Command Code Version

1.29.0

Environment

  • Plan: GOAT
  • Model: DeepSeek V4 Flash
  • Usage Period: August 17–18, 2026
  • Usage Window: All off-peak hours (no peak-time surcharges should apply)
  • ZDR (Zero Data Retention): Not enabled

Operating System

macOS

Terminal/IDE

WezTerm

Shell

zsh

Session file (optional)

No response

Fix prompt (optional)

No response

Additional context

I contacted support and was told that the GOAT plan uses credits and each model consumes them at a different rate. However, this doesn't explain why only Cache Read deviates from the table while Input and Output match perfectly.

I've also discussed this with other users and the hourly pricing theory was suggested (peak/off-peak rates), but since all my usage was during off-peak hours, that doesn't account for the discrepancy.

Note on Trace IDs:

I cannot access the individual trace IDs for those days because the Command Code Studio UI only displays the last 100 requests. I've asked support if there's a way to retrieve the full history, but I don't have the granular per-request data to pinpoint exactly which requests caused the overage. If there's an API endpoint or log export I can use to get the full trace history, please let me know.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions