AI Stats (Astro + Bun)

This is a hobbyist project. Inspired by Theo Browne's model-prices

Pricing, Artificial Analysis benchmark scores, speed, and latency metrics are from Artificial Analysis via their free API. Epoch benchmark runs are from Epoch AI, AI Benchmarking Hub, published online at epoch.ai/benchmarks, used under Creative Commons Attribution licensing with Apache 2.0 licensing for Aider Polyglot and Terminal Bench derived subsets. OpenRouter public model catalog data is used for model availability, provider coverage, usage rankings, provider directory metadata, embedding models, endpoint route detail, modalities, context length, and route pricing metadata. Hugging Face Hub public model data adds public repository popularity and tag signals. LiteLLM public catalog data adds extra provider pricing, context, and capability checks. Public catalog rankings are not treated as benchmark scores.

OpenRouter catalog sync

The Supabase migration in supabase/openrouter_catalog.sql adds:

• public.openrouter_models
• public.openrouter_model_endpoints
• public.sync_openrouter_models()

The dashboard reads from public.openrouter_models when present. If that table has not been applied yet, it falls back to OpenRouter's public /api/v1/models?output_modalities=all endpoint so the catalog collection still renders.

The frontend also uses these free public OpenRouter endpoints when available:

• /api/v1/models/count
• /api/v1/providers
• /api/v1/embeddings/models
• /api/v1/models/{author}/{slug}/endpoints, capped to the top catalog routes
• /rankings/, parsed for public weekly usage signals

The frontend also fetches free public catalog data from the Hugging Face Hub API and LiteLLM's model prices and context catalog. Those sources are used as availability, popularity, pricing, context, and capability signals only. They are not benchmark scores.

Epoch data sync

Epoch AI publishes the free AI Benchmarking Hub export at https://epoch.ai/data/benchmark_data.zip. Apply supabase/epoch_benchmark_sync.sql, then run:

SUPABASE_SERVICE_ROLE_KEY=... bun run sync:epoch

The sync loads the ECI model file and every benchmark CSV in the ZIP. It upserts benchmark definitions by slug, models by model_version, and benchmark runs by Epoch run id while preserving the raw CSV row for source traceability. It also stores every CSV file from the ZIP in public.epoch_data_files, including the auxiliary ECI files that are not direct benchmark leaderboards.

To refresh the frontend fallback snapshot without writing to Supabase, run:

bun run snapshot:epoch

That snapshot is served from public/data/epoch-benchmark-snapshot.json. The dashboard and compare page use it when it has broader Epoch coverage than the hosted Supabase tables, so old Epoch benchmark slugs and newly added free benchmarks stay visible even before the database sync runs.

The frontend promotes the highest-signal free benchmarks first, including GPQA Diamond, FrontierMath, SWE-bench Verified, TerminalBench, Aider Polyglot, HLE, SimpleBench, SimpleQA Verified, ARC-AGI-2, LiveBench, WebDev Arena, CyBench, OSWorld, GDPval, Video-MME, METR Time Horizons, APEX-Agents, The Agent Company, BALROG, DeepResearchBench, and WeirdML. The full Epoch pool still remains available in the metric dropdowns.

Benchmark attribution is shown beside the active metric selector. Artificial Analysis metrics link to Artificial Analysis. Epoch-backed metrics link to the benchmark publisher when the CSV exposes a source link or when the project has a verified fallback, and they also credit Epoch AI Benchmarking Hub as the aggregation source. Unknown future Epoch benchmarks fall back to Epoch AI Benchmarking Hub until a publisher link is available.

The font used is NebulaSans, which is a custom font created by nebula.com all rights reserved to them.

The site is built with Astro and TailwindCSS.

Using Bun

This project is configured to use Bun as the package manager (see package.json packageManager field).

Install Bun (macOS):

brew install oven-sh/bun/bun

Install dependencies (generates bun.lockb):

bun install
cp .env.example .env

Run the dev server:

bun run dev

Build for production:

bun run build

Preview the production build:

bun run preview

Notes:

• If you previously used npm, you can remove package-lock.json and node_modules/ before running bun install to avoid lockfile conflicts.
• Bun can run astro directly as well via bunx astro, but the scripts defined in package.json are recommended.

Citation/Thanks yous: Logo icons are from lobe-icons

License

The source code in this repository is licensed under MIT. Third-party data, logos, font files, and trademarks are excluded. See LICENSE and NOTICE.md.

Data-backed smoke checks

bun run check:epoch validates the Supabase-backed model data path and requires real PUBLIC_SUPABASE_URL and PUBLIC_SUPABASE_ANON_KEY values. In a clean environment without credentials, use bun run lint and bun run build for local portfolio verification.