Pick a question from the categories - the server sends it IN PARALLEL to four free AI models (two from Groq, one from OpenRouter, one from Gemini), and you watch the answers arrive live, word by word, with time and length measurements. Time to first word, speed and answer quality often give very different rankings. The questions are intentionally preset (you can't type free text), so this page can't be abused as an open-ended free AI proxy.
No data yet: run a comparison and cast a vote.
If you'd like to try this exact question yourself, unlimited, with your own key, here's the exact command - copy it, swap in your own key, and run it on your own machine. Your key never reaches us, this request runs directly between your machine and the provider.
A "token" is the basic unit of language models - roughly a word fragment (about 4 characters in English; in Hungarian often fewer per token because of accents and suffixes) - response time and cost are both typically proportional to the number of tokens processed and generated.
"Time to first word" and total response time are two different measures: the first shows when the model starts answering, the second when it finishes. A "thinking" model starts slower but may finish quickly - with live streaming that shows up in the columns immediately.
Separating the system role (system prompt) from the user question keeps the model in the intended role more reliably than cramming everything into one message - a generally accepted pattern in real AI applications as well.
Free API tiers come with quotas and load: if a model is overloaded or retired, the page switches to a fallback model by itself (the column says so), and if a provider drops out entirely, the other columns keep working. Running out of the daily quota is not a bug, it is the providers' business model.
The "auto-check" is only a simple keyword probe (e.g. does the right number appear in the answer): a useful hint, but it does not replace actually reading the answer, and a "correct" end result can come with a hallucinated justification.
This site uses Google Analytics to measure visits. This requires your consent - see the privacy policy for details. Privacy Policy