Claude Code vs Codex: one-shot migration takes 37 min with zero rework vs 27 min but broken twice
notherealironman · reddit · 2026-09-04
The author ran the same refactoring prompt in parallel to migrate a 2000-line Python/Streamlit app to Next.js/TS, comparing Claude Code and Codex:
- Speed: Codex finished in 27 min vs Claude's 37 min
- Code volume: Claude wrote 2x more code, over-engineered and verbose, using 1.5x context/tokens
- Reliability: Claude tested the app in the browser and shipped nearly working on the first pass; Codex required two extra manual prompts to get it running
- UI: Claude nailed padding/spacing/alignment; Codex had more responsive layouts
Takeaway: one-shot prompts aren't rigorous benchmarks, but they expose each agent's trade-offs on long-running tasks — Claude is more hands-off but token-hungry, Codex is faster but needs human intervention.
More from coding & agent
- Your coding agent forgets its rules: 7 paper-backed fixes for compaction loss — victor_explore · 2026-09-04
- Every's consultants answer 33 AI adoption questions from 400 executives — every · 2026-09-04
- Meeting Transcript Extraction Has Plateaued: Where to Improve Quality Beyond Prompt Tweaks — grantnlee · 2026-09-04
- 2,200 Calls, $200 a Day, Zero Cost Attribution: Multi-Agent Run Accounting Has No Standard Answer — mrtrly · 2026-09-04
- Judge at OpenAI's Singapore Sites Hackathon shares how to stand out when everyone builds with agents — gabrielchua · 2026-09-04
- Dev builds local markdown-based long-term memory for AI agents, used daily for 4 months — Rudy_PH · 2026-09-04