OpenRouter gives you one API key for 400+ AI models — GPT, Claude, Gemini, DeepSeek, Llama and more — with automatic fallbacks and pay-as-you-go pricing. With Stripe's $7 billion acquisition putting it in every tech headline this week, here is a complete beginner's tutorial: setup in 5 minutes, your first API call, free models, and the cost-saving tricks most users miss.
Why Use OpenRouter Instead of Direct APIs?
| Problem with direct APIs | OpenRouter's answer |
|---|---|
| Separate accounts/keys for OpenAI, Anthropic, Google... | One account, one key, one bill |
| Provider outage breaks your app | Automatic fallback to another provider |
| Trying a new model = new integration | Change one string: the model name |
| Comparing costs across providers is tedious | Live pricing for every model in one place |
| Some models unavailable in your country | Routed access to nearly everything |
The killer feature is switching cost — or rather, the lack of it. Testing whether DeepSeek does your task at one-tenth of GPT's price becomes a 10-second experiment instead of an afternoon of integration work.
Step 1: Get Your API Key (2 Minutes)
- Go to openrouter.ai and sign up (Google/GitHub login works)
- Open Keys in your account menu → Create Key
- Add credits ($5 is plenty to start) — or skip this and use free models first
Step 2: Your First API Call
OpenRouter is OpenAI-compatible — if you've ever used the OpenAI SDK, you already know how to use it. Just change the base URL and key:
from openai import OpenAI
client = OpenAI(
base_url="https://openrouter.ai/api/v1",
api_key="sk-or-your-key-here",
)
response = client.chat.completions.create(
model="anthropic/claude-sonnet-4.5",
messages=[{"role": "user", "content": "Explain APIs in one paragraph"}]
)
print(response.choices[0].message.content)Want to try a different model? Change one line:
model="deepseek/deepseek-chat" # ~10x cheaper
model="google/gemini-2.5-flash" # fast + cheap
model="openai/gpt-5.6" # OpenAI flagshipThat's the entire migration. Any tool that accepts a custom OpenAI base URL (chatbots, IDE plugins, agent frameworks) works with OpenRouter instantly.
Step 3: Use Free Models (Yes, Really Free)
OpenRouter lists genuinely free models — perfect for prototyping:
- Browse openrouter.ai/models and filter by price = FREE
- Or use the special router
openrouter/free, which auto-selects an available free model - Free tiers have rate limits, so treat them as a sandbox, not production
Smart workflow: build your app on free models, measure quality, then upgrade only the calls that need a premium model.
Step 4: The Cost-Saving Features Most Users Miss
Auto-routing. Use openrouter/auto as the model name and OpenRouter picks a cost-effective model for each request based on your prompt.
Fallback chains. Pass a list of models — if the first is down or rate-limited, the request automatically flows to the next:
extra_body={"models": ["deepseek/deepseek-chat", "google/gemini-2.5-flash"]}Price sorting. In the models list, sort by cost per million tokens. Price differences between models of similar quality are routinely 10–30x — this single habit saves more money than any optimization trick. Provider pricing also changes constantly (OpenAI cut prices 80% this month, while DeepSeek raised theirs 11x) — OpenRouter reflects changes immediately, so re-check your model choices monthly.
Spending limits. Set a hard monthly cap in account settings. Do this on day one — runaway agent loops are the #1 cause of surprise AI bills.
Common Beginner Mistakes
- Hardcoding one model everywhere. Put the model name in a config variable — the whole point of OpenRouter is easy switching.
- Using premium models for simple tasks. Summarizing text or extracting data works fine on models costing pennies; save flagships for hard reasoning.
- Ignoring the Activity tab. It shows exactly which calls cost what — five minutes there usually finds an obvious optimization.
- Exposing your key in frontend code. Like any API key, calls belong on your server, never in browser JavaScript.
Does the Stripe Acquisition Change This Tutorial?
Not yet. OpenRouter has officially confirmed the platform continues operating for existing developers, and nothing in the setup above changes. The sensible precaution is architectural: keep your integration behind a thin wrapper so that if pricing or terms ever shift, switching gateways is a one-line change — exactly the flexibility this tutorial already teaches.
Total setup time: about 5 minutes. Total money saved by being able to choose the right-priced model for every task: typically 50–90% of a naive single-provider bill. Few tools in AI offer that return on so little effort.