Same five jobs, same prompts: the results, the bills, and how to run it yourself

On September 5, 2026 I gave GPT-6 Astra (in Codex CLI) and Claude Fable 5.1 (in Claude Code) the same five everyday jobs with identical prompts and files: build an expense-splitter web app from one prompt, fix a timezone bug that US customers see and London can't reproduce, turn a screenshot of an invoice dashboard into a CSV, find four money clauses planted in a 70,000-word vendor contract, and organize 200 messy files into client and year folders. A checker written before the runs scored every round. Both passed 5 of 5; total time 5:31 vs 5:25; list-price bill $3.29 vs $2.06. This scorecard has the results table, the five prompts word for word, what was in each folder, how the checker scored each round, both itemized bills with the caching explanation, the plan-access table for ChatGPT and Claude (from the vendors' pages), and a 10-minute recipe to run the same test on your own files. Independent test from Hyperautomation Labs.

By submitting, you'll be subscribed to the Hyperautomation AI Report newsletter — three practical AI emails a week. Unsubscribe anytime.

Free. No spam. Unsubscribe anytime.