A curated, searchable database of 610 LLM evaluation benchmarks across 16 capability dimensions — with inline PDF reading, bilingual Draw.io build-process review baselines, bilingual UI, dark mode, neon glow effects, and automated CI/CD.
🌐 Live Demo · 📊 Browse Benchmarks · 🤝 Contribute
| Feature | Costco | PapersWithCode | HuggingFace Datasets | arXiv Search |
|---|---|---|---|---|
| Curated LLM benchmarks only | ✅ | ❌ (all ML) | ❌ (all datasets) | ❌ |
| Inline PDF reading | ✅ | ❌ | ❌ | ❌ |
| Build process flowcharts | ✅ | ❌ | ❌ | ❌ |
| Multi-dim filtering (year/difficulty/openness) | ✅ | Partial | Partial | ❌ |
| Bilingual (EN/ZH) | ✅ | ❌ | ❌ | ❌ |
| Related benchmarks & family lineage | ✅ | ❌ | ❌ | ❌ |
| Dark mode with neon glow effects | ✅ | ❌ | ❌ | ❌ |
| Automated CI/CD deployment | ✅ | ❌ | ❌ | ❌ |
- 610 Benchmarks across 16 capability dimensions — Agent Capability (96), General Language (76), Multimodal (132), Code (63), Science & Reasoning (39), Safety & Alignment (44), Medical & Health (85), and more.
- Neon Glow & Shimmer Effects — Interactive neon glow effect on card hover and a subtle shimmer animation on the logo in dark mode.
- Inline PDF Reading — Click any card to open the details drawer and read the full paper without leaving the page. Most entries embed the original arXiv PDF directly.
- Build Process Flowcharts — All 614 catalog entries include bilingual Draw.io review-baseline diagrams and editable sources, with formal paper-alignment status tracked separately in the manifest and fullscreen mode for complex flows.
- Powerful Filtering — Filter by L1 capability category, year (including 2025/2026 latest), difficulty level (Basic → Frontier), and data openness (Public / Partly / In-house).
- Family & Lineage — Explore benchmark families (e.g., MMLU, GAIA, SWE-bench) and related benchmarks to understand the evaluation landscape.
- Bilingual UI — Full English and Chinese interface with bilingual data fields.
- Automated CI/CD — GitHub Actions automatically validate and deploy updates to GitHub Pages when benchmark data or public assets change.
# Install dependencies
pnpm install
# Local development
pnpm dev
# Build for GitHub Pages
pnpm build:ghpagesThis project uses GitHub Actions for automated deployment. Any push to the main branch that includes changes under client/public/** will trigger validation, a GitHub Pages build, and deployment to the gh-pages branch.
A daily cron job also runs to sync recent public benchmark data and asset changes.
If you need to deploy manually:
- Fork or clone this repository
- Go to Settings → Pages
- Set Source to Deploy from a branch →
gh-pages - Run
pnpm build:ghpages && npx gh-pages -d dist-ghpagesto deploy
Access at: https://<username>.github.io/llm-benchmark-costco/
Sub-path configuration: If deploying under a sub-path, set
base: '/your-repo-name/ 'invite.ghpages.config.ts.
The data lives in client/public/benchmarks.json. Before updating, read CONTRIBUTING.md for the complete workflow covering data schema, validation, and CI process.
| Layer | Technology |
|---|---|
| Frontend | React 19 + TypeScript |
| Styling | Tailwind CSS 4 |
| Build | Vite 7 |
| Routing | Wouter |
| CI/CD | GitHub Actions |
| Icons | Lucide React |
| Diagrams | Draw.io (SVG + editable sources) |
| Deployment | GitHub Pages |
llm-benchmark-costco/
├── .github/workflows/ # GitHub Actions CI/CD
│ ├── ci.yml # PR validation
│ ├── deploy.yml # Deploy on data change
│ └── sync-and-deploy.yml # Daily sync
├── client/
│ ├── public/
│ │ └── benchmarks.json # 610 benchmark entries
│ └── src/
│ ├── components/
│ │ ├── BenchmarkCard.tsx # Card component with neon glow
│ │ ├── BenchmarkDrawer.tsx # Detail drawer + PDF + flowchart
│ │ ├── FilterBar.tsx # Filter controls
│ │ └── Navbar.tsx # Top navigation with logo shimmer
│ ├── contexts/
│ │ └── LangContext.tsx # i18n (EN/ZH)
│ ├── hooks/
│ │ └── useBenchmarks.ts # Data loading & filtering
│ └── types/
│ └── benchmark.ts # TypeScript types
├── scripts/
│ └── validate_benchmarks.py # Data validation script
├── vite.ghpages.config.ts # GitHub Pages build config
└── README.md
We welcome contributions! The easiest way to contribute is to submit a new benchmark via GitHub Issues using the Submit New Benchmark template — no coding required.
For code contributions, please read CONTRIBUTING.md.
MIT
