Get the free Faceoff Protocol
Here's the test everyone argues about but almost nobody actually runs: the $20/mo agent against the free one, same tasks, same machine, stopwatch running. On 2026-07-27 we recorded it. Round 1 (build): one prompt, a typing-speed-test web app in a single file with six scoreable requirements. Claude Code shipped a product-grade app — 566 lines, per-character coloring, keystroke breakdown — in 1m 25s, 6/6. OpenCode, running its free stealth model Big Pickle on the out-of-box Zen tier, shipped a minimal-but-complete 6/6 app in 48.1 seconds — nearly twice as fast, with the cost meter frozen at $0.00 (Claude Code's /cost read $1.33 of API value). Round 2 (debug): a planted boundary bug (> where >= belonged) in an invoicing repo, two tests failing, one instruction: fix it without touching the tests. Claude Code: 18 seconds, one-line fix, ran the tests itself. OpenCode: 22.8 seconds, the IDENTICAL one-line fix. On the work that looks most like your actual day, the free model matched the frontier model fix for fix. This 7-page protocol gives you both exact prompts, the complete planted-bug repo to recreate in two minutes, the 6-point checklist that turns 'it works' into a number, the full receipt table, and the decision guide — including the fine print: OpenCode's free tier is a limited-time beta, free-model data may be used for training (keep private code off it), and Big Pickle is a stealth model nobody can identify. Run the faceoff yourself — it costs nothing to try both.
Free. No spam. Unsubscribe anytime.