# How many real coding tasks fit in one 5-hour window on the $20 plans? I measured it: Sonnet ~33, Haiku ~20, Codex ~18, Opus ~6 — retained record

> Retained reddit record from reddit:ClaudeAI.

Canonical: https://highsignal.app/data/records/0511701f6bfb3a00

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:ClaudeAI)  published 2026-10-24 17:58 ingested 2026-10-25 02:36

[original source ↗](https://www.reddit.com/r/ClaudeAI/comments/1xef9sl/how_many_real_coding_tasks_fit_in_one_5hour/)

## retained text



```
I wanted a straight answer to "what does $20 actually get me" for coding, so I ran the same task on each setup and read the usage meter before and after.

**Setup**
- Task: a real 8-file feature in WeasyPrint (the same one from my earlier token-saver test)
- Every model on its default settings, nothing tuned. Opus runs on medium effort by default in Claude Code, Codex Sol 6.1 on medium in Codex
- Claude Pro ($20) for the Claude models, ChatGPT Plus ($20) for Codex
- Usage read from each tool's own meter, 2–4 runs per setup

**Share of the 5-hour limit one task used**

| setup | per task | tasks per 5h window |
|---|---|---|
| Claude Pro, Sonnet 5.5 | ~3% | ~33 |
| Claude Pro, Haiku 5.5 | ~5% | ~20 |
| ChatGPT Plus, Codex Sol 6.1 | ~5.5% | ~18 |
| Claude Pro, Opus 5.5 | ~15% | ~6 |

Weekly limit barely moved per task: about 1% for Opus, about 0.5% for Sonnet and Codex.

**The surprise:** Haiku is the "cheap" model and still fit fewer tasks than Sonnet. My guess is it took longer and went back over the code more to get the same feature done, so it ate more of the window even at a lower rate.

**Caveats**
- 2–4 runs per setup, so these are estimates, not exact numbers
- One task type. A small bug fix or a huge refactor would shift the numbers
- The meter only shows whole percentages, so the small ones (3%, 5%) are rough
- Defaults only. Opus on low effort costs a lot less; I tested that separately and will post it

If you're on Pro and mostly writing features, Sonnet gets you the most work per window by a wide margin.
```



document fetched 2026-10-25 00:21 · canonical https://reddit.com/r/ClaudeAI/comments/1xef9sl/how_many_real_coding_tasks_fit_in_one_5hour
