---
title: "Discussion on \"36 Days of Claude Code Logs: Silent Model Switching, 11.5x Efficiency Gap\" | Hashnode"
description: "Discussion on \"36 Days of Claude Code Logs: Silent Model Switching, 11.5x Efficiency Gap\". Continuing the cache TTL audit series, this time breaking down by model. The server silently switched main agent models three times (Opus 4.6 → Sonnet 4.6 → Opus 4.7 → Sonnet 4.6), and Claude Code autonomously assigns sub-agent models. Comparing cost per million main output tokens across 7 periods reveals an 11.5x efficiency gap."
url: "https://hashnode.com/posts/claude-code-model-cost-efficiency/69fef38ecb5b68a0fa1dbbfd"
---

[Feed](https://hashnode.com/feed) /

Discussion

# [36 Days of Claude Code Logs: Silent Model Switching, 11.5x Efficiency Gap](https://recca0120.hashnode.dev/claude-code-model-cost-efficiency?utm_source=hashnode&utm_medium=feed)

Continuing the cache TTL audit series, this time breaking down by model. The server silently switched main agent models three times (Opus 4.6 → Sonnet 4.6 → Opus 4.7 → Sonnet 4.6), and Claude Code autonomously assigns sub-agent models. Comparing cost per million main output tokens across 7 periods reveals an 11.5x efficiency gap.

May 9 · 9 min read · recca0120.hashnode.dev  
Read the article

-   [#ai-agent](https://hashnode.com/tag/ai-agent)
-   [#claude-code](https://hashnode.com/tag/claude-code)
-   [#prompt-caching](https://hashnode.com/tag/prompt-caching)
-   [#python](https://hashnode.com/tag/python)

## Written by

[recca0120](https://hashnode.com/@recca0120)

@recca0120 · Taipei, Taiwan

  
[View profile](https://hashnode.com/@recca0120)

## Responses

Markdown supported · ⌘↵ to post

No responses yet.