The gallery
Six coding agents got the same three open briefs. No score, no right answer — this is what each one built when the shape was left to it. Every workspace below is the real thing.
Full results and method: notionbench.rishabhsai.com
A study system
Spaced repetition, courses, notes, weak spots, assigned reading.
- Claude Code × Opus 5 — 8 databases, 46 rows, 9 views
- Claude Code × Fable 5 — 8 databases, 88 rows, 9 views
- Claude Code × Sonnet 5 — 2 databases, 0 rows, 2 views
- Codex × Sol (xhigh) — 9 databases, 21 rows, 10 views
- Codex × Luna (high) — 5 databases, 5 rows, 5 views
- Codex × Sol (medium) — 4 databases, 5 rows, 4 views
A mobile grooming business
Recurring bookings across two vans, supplies, servicing, and whether van two pays for itself.
- Claude Code × Opus 5 — 11 databases, 1,633 rows, 14 views
- Claude Code × Fable 5 — 9 databases, 103 rows, 9 views
- Claude Code × Sonnet 5 — 9 databases, 47 rows, 9 views
- Codex × Sol (xhigh) — 10 databases, 18 rows, 14 views
- Codex × Luna (high) — 5 databases, 5 rows, 12 views
- Codex × Sol (medium) — 6 databases, 10 rows, 6 views
A Twitch creator's pipeline