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
A user measured the number of coding tasks achievable within a 5-hour window on various $20 AI plans. Claude Pro's Sonnet 5.5 completed approximately 33 tasks, using about 3% of the weekly limit per task. Haiku 5.5 managed around 20 tasks, while ChatGPT Plus's Codex Sol 6.1 completed about 18 tasks. Claude Pro's Opus 5.5 performed the fewest at roughly 6 tasks, consuming about 15% of the weekly limit per task. Sonnet 5.5 offered the most work per window for feature writing.
This report uniquely quantifies the actual coding tasks completed per 5-hour window across different $20 AI plans, unlike other comparisons that focus on feature sets or raw performance metrics.
Time & source
Times shown in UTC
Display time zone: UTC
Local time zone unavailable; showing UTC.
PublishedOffset at this time: UTC+0Oct 24, 2026, 17:58 UTC
IngestedOffset at this time: UTC+0Oct 24, 2026, 21:00 UTC
- Published
- Oct 24, 2026, 17:58
- Ingested
- Oct 24, 2026, 21:00
- Source type
- Dev community
- Tier
- Community
- Source status
- Healthy
Tier is a per-source editorial setting, not a per-item score.
Discussion trend
The percentage is based on collected discussion signal, not new comments or independent people. The curve only compares the same topic across time.
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
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.