Plansmith
PlanSmith Benchmark

Accounting & invoicing — the model build-off

We dogfood every PlanSmith planner across multiple coding models. We lock the vertical, the discovery answers, and the feature ledger, then audit what each one builds. The planner is the product — these are the apps it produced. Every one is a working double-entry bookkeeping app; the ranking is a hands-on read of each raw build, not a number.

2026-07-063 models · 7 builds audited
The build-off · 01

How the builds ranked.

01

Penny

Best Overall
by Claude Opus 4.8

Best for a deep, buyer-usable accounting build.

Best at
The most complete of the set — proper double-entry bookkeeping across multiple companies you can switch between, with invoices, expenses, supplier bills, bank reconciliation, and profit-and-loss and balance-sheet reports, over demo books rich enough to fill every dashboard panel.
Watch for
Only deployment awkwardness — nothing a user would see or feel.
Open live demo
02

Ledgerly

Cleanest Handover
by Claude Opus 4.8

The safest, most disciplined build to hand over.

Best at
Built for the person actually keeping the books: an approval queue for supplier bills, a bank-reconciliation workspace, and reports on who owes you and who you owe, with permissions checked on the server for every single action.
Watch for
The interface reloads pages the traditional way rather than feeling like a slick modern app.
Open live demo
03

Beginner-friendly ledger

Most Approachable
by Claude Opus 4.8

Best for a friendly, first-timer accounting app.

Best at
The whole bookkeeping stack — chart of accounts, journal entries you can reverse, trial balance, balance sheet and profit-and-loss — behind a warm, guided setup and plain-English wording throughout.
Watch for
It would not finish building for production, and once hosted it sent people to the wrong addresses. Both only appeared once it left a developer's machine.Fixed since auditit now builds cleanly, sends people to the right pages once hosted, signs in with one click, and arrives with the books filled in.
Open live demo
04

Bookkeeper Finance OS

Most Ambitious
by GPT-5.5

Widest ambition; worth a review before handover.

Best at
The most ambitious scope — several companies, several currencies, an approval workflow, reconciliation, a tax summary and a complete record of who changed what. It also carries its own database, so there is nothing to set up.
Watch for
It arrives empty — no transactions at all, so every ledger is blank on first open — and left its own test scratch-work lying around.Fixed since auditthe demo is seeded with a full set of books, so the dashboard lands populated.
Open live demo
05

MintBooks

Focused & Lean
by GPT-5.5

Best when a narrow, invoice-led scope is enough.

Best at
Lean and invoice-led, running on nothing but itself — cash accounts, a running ledger, invoices that accept part payments, receipts, profit-and-loss, and a view of who owes you and for how long. Its own tests check the maths to the exact cent.
Watch for
Deliberately narrow: no chart of accounts, general ledger, balance sheet or bank reconciliation. Its demo login also refused to work once hosted.Fixed since auditthe one-click demo login now works on the hosted demo.
Open live demo
06

Butter

Underrated · Best UI
by Claude Opus 4.8

Stronger than its audit rank suggests.

Best at
34 features over books that always balance, behind the cleanest and plainest interface of the set — invoices, expenses, bank matching, the general ledger and profit-and-loss — each one hardened against a deliberately hostile security review.
Watch for
It scored badly for looking broken at sign-in, but that was a hosting quirk rather than a fault in the app, alongside a security review it had already mostly acted on and disclosed.Fixed since auditsign-in now works on the hosted demo with one click, and the books arrive filled in.
Open live demo
07

Meridian

Late Entry · Solid Core
by Gemini 3.5 Flash

Newest build — beat the “too shallow” expectation.

Best at
Real double-entry bookkeeping in one place — general ledger, trial balance, balance sheet, profit-and-loss, and bank reconciliation that flags anything that does not match — over books that arrive already populated.
Watch for
It arrived after the original audit, and a couple of its report tabs shipped as “coming soon” placeholders.Fixed since auditthe Profit & Loss and Balance Sheet reports now render in full.
Open live demo

Models tested: Claude Opus 4.8, GPT-5.5, and Gemini 3.5 Flash. The four Claude Opus 4.8 entries are the same model across different build lanes — not different model versions. This is a PlanSmith dogfood benchmark for specific accounting-invoicing runs, not a universal ranking of coding models; each run used the planner package and discovery answers available at the time (which were not fully scope-identical). Ranks reflect each build at audit time. The live demos have since been optimized for presentation — improvements are marked “fixed since audit.”