Model leaderboard

Which model wins, where

Not a single number: the leaderboard splits every recorded run by industry and difficulty, so you can see whether Claude owns marketing or GLM owns dashboards. All scores come from the live run log, not editorial opinions.

models tracked

88%

top overall

99%

most stable

Model leaderboard — sorted by run volume, broken down by the industry each model wins and loses in
Model	Runs	Overall	Wins in	Wins avg	Loses in	Loses avg	Error rate
01GLM-4.6 (CN)	293	88%	Agency	92%	Local business	81%	8 of 293
02Claude 4.6 Sonnet	184	90%	A11yStrict	94%	Utility	83%	1 of 184
03Codex	184	88%	AI	95%	Food	85%	8 of 184

By difficulty — the hard question: which model do you trust when the brief gets gnarly
Model	Starter	Intermediate	Advanced	Runs
GLM-4.6 (CN)	88%	86%	90%	293
Claude 4.6 Sonnet	90%	89%	94%	184
Codex	88%	87%	91%	184

Briefs that fit

Pick the model you actually use, the industry you're shipping in, and the difficulty of the brief — this shows which prompts from the catalog that combo will handle best.

ModelIndustryDifficulty

184 briefs match this combo · sorted by measured fidelity

AI tool landing with terminal motifAI

technical, confident

avg 93%best Codex4 runs

Dark SaaS launch page with aurora heroSaaS

dark, premium, minimal

avg 92%best Claude 4.6 Sonnet4 runs

Rewilding campaign pageNonprofit

earnest, hopeful, concrete

avg 92%best Claude 4.6 Sonnet4 runs

A11y-strict page — WCAG AA as a hard requirementA11yStrict

spec-exact, keyboard-clean, audit-visible

avg 92%best Claude 4.6 Sonnet4 runs

Agency site with editorial case studiesAgency

editorial, confident

avg 91%best GLM-4.6 (CN)4 runs

Law firm that reads like a personLegal

credible, calm, plain-spoken

avg 91%best GLM-4.6 (CN)4 runs

Indie magazine — issue 09 landingPublishing

editorial, tactile, playful

avg 91%best GLM-4.6 (CN)4 runs

Dev-tool API reference with try-it paneDevTools

developer-calm, precise, searchable

avg 91%best Claude 4.6 Sonnet4 runs

Boutique hotel — real amenitiesHospitality

linen-calm, honest, neighbourhood

avg 91%best Claude 4.6 Sonnet4 runs

Scores are a report card, not a benchmark. Fidelity here is an editorial judgement recorded per run — the same numbers a /prompts card shows, just re-projected to answer the “where should I spend my time” question. Every run log on this site says so too.