What for: one key and one bill → hundreds of models (Claude, GPT, Gemini, open ones). No zoo of separate accounts and billings.

What it is and who it's for

OpenRouter is a single gateway to hundreds of LLMs from different providers through one API key and one bill. For the developer, product manager or researcher who doesn't want separate accounts and billing setups at OpenAI, Anthropic, Google and a dozen open-model hosts. What sets it apart: it speaks the OpenAI-compatible format, so code written against the OpenAI SDK works with almost no changes, and swapping models is a one-line edit.

Key features

One endpoint and one key for Claude, GPT, Gemini, Llama, DeepSeek, Qwen and hundreds of other models.
Automatic fallback and routing: if a provider goes down or gets expensive, the request goes to a backup.
Per-token price comparison between providers of the same model right in the catalog.
BYOK (bring your own provider key) — you can push traffic through your own account and let OpenRouter just do the routing.
Unified spend tracking, limits and stats per app/key.
Free-tier models (marked :free) for trying things out without paying.

How to start in 5 minutes

Sign up at openrouter.ai, top up your balance (a couple of dollars is enough) and create a key under Keys.
Point your code at the base URL and the key — the format is the same as OpenAI's:

curl https://openrouter.ai/api/v1/chat/completions \
  -H "Authorization: Bearer $OPENROUTER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"anthropic/claude-3.5-sonnet","messages":[{"role":"user","content":"Hello"}]}'

Change model to any slug from the catalog — the rest of the code stays untouched.

When to take it, when not to

✅ Take it if: you need access to MANY models at once, A/B testing across providers, fallbacks and one bill instead of a zoo of accounts.
✅ Take it if: you're building a product and want to switch models per task without rewriting the integration.
❌ Don't take it → "Vercel AI Gateway" is better if you're already on Vercel and want BYOK with no markup and per-project budgets.
❌ Don't take it → "Ollama" or "LM Studio" is better if privacy and offline use matter — here your data still goes to the provider's cloud.

The honest price

The gateway itself: you pay per token at the providers' rates plus a small fee on top-ups; there are free :free models with hard rate limits. Exact numbers are on the provider's side and change over time.

Gotchas

Free models often hit rate limits and can change availability without warning.
Latency and uptime depend on the provider actually serving the request, not on OpenRouter.
Not every provider supports the full feature set (vision, tools, streaming) equally — check per model.
Confidential data still leaves your perimeter — for PII and secrets you need self-hosting.

🤖 Prompt accelerator

"I'm using OpenRouter (OpenAI-compatible API). Help me set up routing for this task: <DESCRIBE THE TASK — chatbot / summarization / code>. Pick 2-3 models by the price/quality/speed balance, give me the model slugs, a sample request in <LANGUAGE: Python/Node>, and a fallback scheme: primary → backup on error or when it gets too expensive. Budget per request ≈ ."