Eleven models, a picker and a no-key playground

Most model catalogues ask you to trust a ranking. Ours now lets you check one. This week the work was to put the catalogue, the choice and the first answer in three places a visitor can reach without an account.

What you can do now

• See every row. The model pages now cover all eleven catalogue rows. Each page states the context window and the input and output price per million tokens, so a 256K window is a number you can compare instead of a label.
• Choose by constraint. The picker indexes the same eleven rows by five questions: which rows fit at least 256K of context (five do), which reach a million tokens (two do), the cheapest input price (seven cents per million), the cheapest output price (thirty cents per million), and the one row the free tier covers. Every value is read from the catalogue the API serves, not typed in by hand.
• Try one prompt without a key. The playground answers a single prompt on the free model with no account and no key. The answer is capped at 512 tokens, and when a limit is reached the page shows the API's own message instead of hiding it.
• Compare three with your own key. Paste your own key and the same prompt runs on three models at once: the free one, a cheap mainstream one and one with a million-token window. Each answer or error lands in its own column, so one failure does not blank the others.
• Share the question, not the key. A share link carries the prompt and the three model names only. It never carries a key, so whoever opens it runs it with their own.

What that means for you

If you have been choosing a model by reputation, the picker now gives you five facts and no ranking. That is deliberate: the constraint is yours to apply. If you only want to see how an answer reads, the free path costs nothing and asks for nothing. If you already hold credits, the comparison is the fastest way to see how the same question lands on three different models before you commit a workload.

Each paid answer is capped at 512 tokens, so a comparison is a sample, not a benchmark. The page says so, and it also says that an extension running in your browser can read any key you type.

What we are not claiming

The picker does not rank models, publish benchmark scores, or say which one is best for a task. It repeats the catalogue's numbers and lets the constraint decide. Prices move with the upstream catalogue, so the live page is the source of truth rather than this post. The playground keeps no history and saves no conversation; a shared link restores the question, not the answer.

The same principle applies to this page. It is a record of what a visitor can do now, not a promise about what comes next.