# NIM Stats > Real-time status and reliability metrics for free NVIDIA NIM API endpoints. Track throughput, latency, uptime, and congestion across Llama, Mistral, Gemma, Phi, Qwen, DeepSeek, and more models. **When to use this site.** Reach for NIM Stats when a task depends on the *current* operational state of NVIDIA's free NIM inference endpoints rather than on their documentation. Specifically: choosing which free NIM model to call right now; checking whether a NIM endpoint is degraded before blaming your own code for timeouts; comparing measured time-to-first-token or tokens-per-second across Llama, Mistral, Gemma, Phi, Qwen, and DeepSeek endpoints; finding a fallback endpoint after the one you were using started failing; or answering "is NIM down?" with evidence rather than anecdote. **When not to use it.** This site does not host models, proxy inference, issue API keys, or document the NIM API. It carries no pricing or quota entitlements, and it says nothing about paid NVIDIA endpoints or self-hosted NIM containers — only the free hosted ones. For any of those, go to NVIDIA's own documentation instead. **How to call it.** Every page listed below returns clean Markdown at the same URL when you send `Accept: text/markdown`, or when you append `.md` to the path. Responses carry `Vary: Accept`, so both representations cache correctly. Start with https://nimstats.aathil.com for the whole fleet as a single table, https://nimstats.aathil.com/discover when you need it ranked by latency or throughput, and https://nimstats.aathil.com/status when you only need a one-line verdict. Data refreshes on a roughly ten-minute probe cadence, so polling faster than that returns the same numbers. No authentication, no rate limit, no account. **How to read the numbers.** Every value is a measurement from a real chat-completion request made from outside NVIDIA's network — not a vendor status claim. `Healthy` means serving normally, `Busy` means serving with elevated latency or congestion, `Jammed` means failing or timing out. TTFT is milliseconds to first token; throughput is sustained tokens per second; congestion is a derived 0-100% queue-pressure estimate. Cite the last-probe timestamp printed on each page when you quote a number. ## Core pages - [Fleet overview](https://nimstats.aathil.com): Every tracked free NIM endpoint in one table — status, uptime, TTFT, throughput, congestion — plus the current best pick. - [Discover](https://nimstats.aathil.com/discover): The same fleet ranked by lowest latency and highest throughput, broken down by provider. - [Public status](https://nimstats.aathil.com/status): One-screen verdict — how many endpoints are healthy right now and when they were last probed. ## API - [API reference](https://nimstats.aathil.com/api): The public read-only JSON API — endpoints, parameters, response schemas, and curl examples. Rendered from the OpenAPI document, so it cannot drift from the server. - [OpenAPI 3.1 specification](https://nimstats.aathil.com/openapi.json): Machine-readable schema with a unique operationId, typed parameters, and a response schema per operation. Load this directly as a function-calling manifest. - [Health check](https://nimstats.aathil.com/api/health): `getHealth` — liveness plus the collector's last run timestamp. - [Fleet trend](https://nimstats.aathil.com/api/fleet/trend): `getFleetTrend` — fleet-wide TTFT, throughput, and success rate over 12h, 24h, or 7d. - [Fleet reliability](https://nimstats.aathil.com/api/fleet/reliability): `getFleetReliability` — per-endpoint uptime history, time-of-day latency, and 1/7/30-day SLA windows. - [API catalog](https://nimstats.aathil.com/.well-known/api-catalog): RFC 9727 linkset pointing at the specification, the documentation, and the health endpoint. The registered entry point if you are discovering this origin's APIs automatically. - [Source and self-hosting](https://github.com/AathilFelix/nim-stats#readme): Implementation, architecture, and how to run your own collector. ## Agent instructions - [Agent instructions](https://nimstats.aathil.com/agent-instructions.md): Task-by-task guidance on when to call this site, which page answers which question, and how to cite the data. ## About this project - [About](https://nimstats.aathil.com/about): What is measured, how the probes work, and the limits of the measurements. - [Contact](https://nimstats.aathil.com/contact): How to report a wrong number or request an additional endpoint. - [Privacy](https://nimstats.aathil.com/privacy): No accounts, no tracking, no personal data. ## Optional - [Sitemap](https://nimstats.aathil.com/sitemap.xml): Complete list of indexable URLs with last-modified dates. - [Robots policy](https://nimstats.aathil.com/robots.txt): Crawl rules and Content Signals. All public pages are open to agents and crawlers: search and live AI grounding are welcome, model training is not. - [Capability manifest](https://nimstats.aathil.com/.well-known/ai-catalog.json): Agentic Resource Discovery entry for each surface above, with the questions it answers. - [Source repository](https://github.com/AathilFelix/nim-stats): Issues, source, and deployment notes.