--- title: inference-metrics emoji: 🐳 colorFrom: blue colorTo: green sdk: static pinned: false --- # LLM Pricing A tool to fetch and compare LLM pricing and capabilities across multiple providers. ## Data Sources This tool uses two primary data sources: 1. **HuggingFace Router API** (https://router.huggingface.co/v1/models) - Primary source for model pricing, context length, and capability flags 2. **Provider-specific APIs** - Fallback source for additional metadata and capabilities The HuggingFace Router API now provides comprehensive data including: - Pricing (input/output costs per million tokens) - Context length - supports_tools flag - supports_structured_output flag - Provider status When data is available from both sources, the HuggingFace Router data takes priority. ## Installation ```bash bun install ``` ## Usage ```bash # Fetch all models and enrich with provider data bun run get-metrics.ts # Skip specific providers bun run get-metrics.ts --skip-providers novita featherless # Test performance for models (requires HF_TOKEN) HF_TOKEN=your_token bun run get-metrics.ts --test-performance # Test specific number of models HF_TOKEN=your_token bun run get-metrics.ts --test-performance --test-limit 10 ``` ## Supported Providers - **novita** - Full API support - **sambanova** - Full API support - **groq** - Full API support - **featherless** - Full API support - **together** - Full API support - **cohere** - Full API support - **fireworks** - Full API support - **nebius** - HF Router data only - **hyperbolic** - HF Router data only - **cerebras** - HF Router data only - **nscale** - HF Router data only ## Output Files - `enriched_models.json` - Complete enriched model data - `provider_models_raw.json` - Raw provider API responses for debugging ## Environment Variables Optional API keys for fetching provider-specific data: - `NOVITA_API_KEY` - `SAMBANOVA_API_KEY` - `GROQ_API_KEY` - `FEATHERLESS_API_KEY` - `TOGETHER_API_KEY` - `COHERE_API_KEY` - `FIREWORKS_API_KEY` - `HF_TOKEN` - Required for performance testing This project was created using `bun init` in bun v1.2.4. [Bun](https://bun.sh) is a fast all-in-one JavaScript runtime.