Real votes and real per-call numbers, time to first token, tokens per second, total tokens, build an honest leaderboard of which model is actually worth using.
Start a threadChoose from OpenRouter's live free-tier catalog, sorted by context window. Defaults to three, swap any of them out.
Every selected model answers at once, streaming independently — one being slow or down never blocks the others.
Once two or more models have answered, pick the best one. That's one real vote, feeding an honest leaderboard.
Most model comparisons are anecdotal. OpenArena keeps the actual measurements next to every answer, and the leaderboard is built from real head-to-head votes, never a made-up score.
Measured in the response itself, not estimated — the real wall-clock gap before the first word arrives.
Computed from real output and real duration for every answer, every time.
Every model here is free tier. That's not a bug or a rounding error — it's shown because it's true.
Browse the live leaderboard, or the full free-model catalog, before you start your own thread.