---
title: "Discussion on \"What a Coding-Agent Session Really Costs: Estimating Spend and Cache Hits from a /models Pricing Feed\" | Hashnode"
description: "Discussion on \"What a Coding-Agent Session Really Costs: Estimating Spend and Cache Hits from a /models Pricing Feed\". Coding agents like Claude Code, Codex CLI, and opencode don't send \"a prompt\". Every turn re-sends the system prompt, the tool schemas, and the whole conversation so far. A 40-turn session can easily"
url: "https://hashnode.com/posts/what-a-coding-agent-session-really-costs-estimating-spend-and-cache-hits-from-a-models-pricing-feed/6aeacbd62d48859dc446833b"
---

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

Discussion

# [What a Coding-Agent Session Really Costs: Estimating Spend and Cache Hits from a /models Pricing Feed](https://wokey.hashnode.dev/what-a-coding-agent-session-really-costs-estimating-spend-and-cache-hits-from-a-models-pricing-feed?utm_source=hashnode&utm_medium=feed)

Coding agents like Claude Code, Codex CLI, and opencode don't send "a prompt". Every turn re-sends the system prompt, the tool schemas, and the whole conversation so far. A 40-turn session can easily

Nov 4 · 8 min read · wokey.hashnode.dev  
Read the article

-   [#python](https://hashnode.com/tag/python)
-   [#llm](https://hashnode.com/tag/llm)
-   [#ai](https://hashnode.com/tag/ai)
-   [#api](https://hashnode.com/tag/api)

## Written by

[Wokey](https://hashnode.com/@wokey)

@wokey

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

Low-cost Claude & GPT API with offline-verifiable responses.

## Responses

Markdown supported · ⌘↵ to post

No responses yet.