Model Leaderboard
Every figure as recorded.
You reduce LLM latency and time to first token mainly by choosing a faster model for the request: time to first token is set by the model and provider serving it. This leaderboard ranks models by recorded p50 latency next to quality and price, so you can shortlist the fastest model that still fits the task — then route to it through the gateway.
Methodology.
Benchmark scores and latency on this page come from archived source records whose measurement conditions are not recorded. A price marked “list price” is the current list price of a model nRouter serves; see the Model Directory for current availability and pricing.
Sources: https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset (updated Oct 19, 2026)
