This site’s method never changes: your cases are worth more than leaderboards. This guide takes you from having no account to watching three models answer the same question of yours, side by side, in a quarter of an hour. The first four steps happen in the browser; the terminal only appears in the last, optional one. If you first want to understand what OpenRouter is and what it costs, start from the tool sheet; here we do.

Step 1 · The account

Go to openrouter.ai and sign up (email or a Google/GitHub account). The account is free, and already lets you use the models with the :free suffix in their name: the cap is 50 requests a day, enough to sniff around the interface, tight for real work.

Step 2 · The cap before the money

Before loading credits, decide how much you are willing to spend: that is the right order, because a cap set afterwards is a cap you have already eaten through. A one-time ten dollars covers weeks of testing (a three-model comparison costs cents) and unlocks a permanent bonus: the :free daily limit rises from 50 to 1,000 requests (checked on 19/07/2026 against the documented limits). OpenRouter applies a 5.5% fee when you buy credits; model usage is charged at list price, with no markup.

Step 3 · The key, on a leash

In your account settings create an API key and give it a spending limit with a monthly reset, say $5. If a script goes wild or the key ends up where it shouldn’t, the maximum damage is the number you wrote. The key is shown once: keep it in a password manager, never in a shared document and never inside a file that ends up in a repository.

Step 4 · Three models, one case of yours

Open openrouter.ai/chat and add three models to the same conversation: a large one, a mid one, a small one (the comparison holds up to five). Then paste a real case from your work, with the context you would give a colleague: the difficult email to rewrite, the minutes to summarise, the table to interpret. Look at the answers side by side and ask yourself which one you would have had to correct least. Repeat with two or three more cases: patterns that hold across different cases are worth more than any leaderboard.

Step 5 (optional) · The structured comparison

If the terminal doesn’t scare you, the course’s test bench makes the comparison repeatable: download esegui.mjs and casi.json, and first watch it work with no key and no cost:

node esegui.mjs --prova

Trial mode uses fake answers and one case fails on purpose, to show you how to read a failure. Then set your key and do the real run:

export OPENROUTER_API_KEY=sk-or-...
node esegui.mjs

Replace the cases in casi.json with 5-10 real inputs from your work and change the model in the MODELLO line to compare: same set of cases, different models, comparable results.

The typical mistakes

The key in the wrong place. A key pasted into a shared document or committed to a repository is a credit card left on the table. The spending limit from step 3 reduces the damage, the password manager avoids it.

The model that is no longer there. The catalogue moves fast: a name copied from a three-month-old article may be deprecated. Always take the exact name from the live price list, where every model shows its slug next to its price.

The :free pace. Besides the daily cap there is a per-minute limit: a hammering script gets 429 errors. For sustained use, small paid models cost very little anyway.

---

Testing status: steps and figures checked against OpenRouter’s official documentation on 19/07/2026. The real run of step 5 was executed on the same date, with a real key and real spend, by the site’s testing workflow: 5 cases against a frontier model, 3 passed and 2 failed. The failures are the bench doing its job: they show where the sample instructions need fixing, which is exactly what you will see on your own cases.