Skip to content
The0pxPublic

About

Open-source AI model comparator: benchmarks, speed, pricing, context windows and a coding-agents leaderboard. FR/EN.

Resources

Security policy

Stars

5 stars

Watchers

0 watching

Forks

Latest commit

 

History

168 Commits

Folders and files

Repository files navigation

BenchSift

Live site: benchsift.nxtaigen.com License: MIT

The BenchSift home page: the AI model ranking with score, price, speed and context

BenchSift brings Artificial Analysis, OpenRouter, Models.dev and Hugging Face data together in one open-source comparator, in English and French, with a coding-agent ranking that shows each agent's cost and time per task next to its score.

BenchSift helps you compare AI models using text and media benchmarks, speed, pricing, and context windows. Open it at benchsift.nxtaigen.com (version française).

Data comes from Artificial Analysis, OpenRouter, and Hugging Face. Refresh frequency depends on the deployment schedule; see DOKPLOY.md.

Features

  • Model catalogue — browse text, image, voice, video and utility models with key stats at a glance
  • Side-by-side comparison — pick up to several models and compare them on every metric
  • Detailed model pages — context window, output speed, pricing, text benchmarks (MMLU, HumanEval, MATH…) and media ELO benchmarks when Artificial Analysis exposes them
  • Search & filter — find models by name or provider instantly
  • Light/dark theme — persisted across sessions
  • French & English — every page has its own address in each language (/models, /fr/models); the browser language is only suggested, never forced
  • No account required — language, theme and comparison preferences stay in the browser
  • Indexable comparisons — static pages such as /compare/<model-a>-vs-<model-b> for the week's most used models, with a summary written from the data

Getting started

Use the Bun version declared in package.json (also used by Dockerfile).

bun install
cp .env.example .env   # then fill in your API keys
bun run dev

Then open http://localhost:3000

You'll need an Artificial Analysis API key — create one from the Data API and add it to .env:

ARTIFICIAL_ANALYSIS_API_KEY=your_key_here
OPENROUTER_API_KEY=your_openrouter_key_here

OpenRouter is used for model metadata, weekly usage rankings and benchmark enrichment. OPENROUTER_API_KEY is optional, but enables authenticated /api/v1/models requests instead of relying on the unauthenticated public path. Hugging Face is used only for official model repository metadata and links. The Artificial Analysis integration uses the supported V2 endpoints, paginates the complete language-model catalogue, and automatically uses the richer Pro or Commercial response when the configured key grants access.

Project structure

src/
  router.tsx              # Router instance, with the /fr URL rewrite
  routes/
    __root.tsx            # Root layout, <head>, providers, error boundary
    index.tsx             # Homepage — catalogue and rankings
    compare.tsx           # Side-by-side comparison
    compare_.$pair.tsx    # Static pair pages, /compare/<a>-vs-<b>
    agents/coding.tsx     # Coding-agents leaderboard
    models/$slug.tsx      # Model detail page
    api/cron/refresh.ts   # Authenticated cache-refresh endpoint
    api/cron/status.ts    # Authenticated cache summary
    health.ts            # Public liveness and catalogue state
    robots[.]txt.ts       # robots.txt
    sitemap[.]xml.ts      # sitemap.xml
  styles/globals.css      # Tailwind v4 entry
components/               # UI components (shadcn/ui based)
lib/
  api.ts                  # Catalogue ingestion and enrichment — server only
  server-fns.ts           # TanStack Start server functions (route loaders)
  revalidate-cache.ts     # Revalidating in-memory cache
  cron-cache.ts           # Persisted models cache for Bun/Dokploy
  home-catalog.ts         # Lightweight homepage payload
  model-reasoning.ts      # Reasoning families and Normal-mode collapse
  provider-map.ts         # Canonical creators/providers
  openrouter-model-filter.ts # OpenRouter model exclusions
  coding-agents.ts        # Coding-agent types + harness metadata (client-safe)
  i18n.tsx                # French/English translations
  locale.ts               # Language in the URL (/fr prefix) and hreflang paths
  seo.ts, route-seo.ts    # Head tags, hreflang, BreadcrumbList and Dataset
  data-summary.ts         # Data-generated summaries (leader, value, fastest)
  compare-pairs.ts        # Popular comparison pairs
  compare-store.tsx       # Client-side comparison state

Built with

Validation and contribution

AGENTS.md defines repository contracts and checks by change type. PRODUCT.md describes the intended UI and interaction behavior. Use SECURITY.md to report vulnerabilities privately.

Deploying

Build and run the production server locally with:

bun run build
bun run start

start runs bun .output/server/index.mjs. The included Dockerfile packages this service for Dokploy. See DOKPLOY.md for environment variables, persistent storage, scheduling, and deployment verification.

About

Open-source AI model comparator: benchmarks, speed, pricing, context windows and a coding-agents leaderboard. FR/EN.

Resources

Security policy

Stars

5 stars

Watchers

0 watching

Forks

Releases

Packages

Used by

Contributors

Languages