A leaderboard tells you how models do on our tasks. The only test that matters is yours. The playground run page takes any prompt and sends it to the models you choose; there are three ways to pay for that, and beginners should start with the one that costs nothing.
| Mode | What it runs | What it costs you | Best for |
|---|---|---|---|
| Keyless | $0 rows and the cheapest model per family | Nothing; three runs per visitor per day | First tests, and proving a cheap model is enough |
| Your own OpenRouter key | Any model on OpenRouter | Your key's balance, at catalogue rates; a typical short run is a fraction of a cent | Testing the exact model you plan to use |
| Pro | The whole catalogue on the site's key, results saved | The Pro subscription; runs draw from a monthly credit | Repeating a test across many families without managing keys |
Writing a prompt that gives a pass/fail answer
- 1
State the output shape in the first line
"Return one JSON object with keys…" or "Output three lines, nothing else." Vague prompts produce answers that are impossible to score.
- 2
Put the facts in the prompt
Paste the text the model should work from. A self-contained prompt gives every model the same information, which is the whole point.
- 3
Add two or three checks
A
containsfor a fact that must appear, anot_containsfor one that must not, amax_wordsfor length. Checks turn 'looks fine' into a badge. - 4
Run it keyless first
If the cheapest model passes every check, you have your answer and it cost nothing. Only reach for a key if it fails.
Try it
One-sentence release note
Three badges: length, the required phrase, and a regex that fails if there is more than one sentence. When all three are green on a keyless model, note the cost line; it will be a fraction of a cent. That is the number you will compare against the flagship in the next lesson.
Knowledge check
You want to know whether the cheapest model can extract five fields from your invoices as JSON. Which mode should you try first, and why?