Speed: output tokens per second

Questions people ask

What is the fastest LLM in tokens per second?

On 24 Oct 2026, Celeris-1 by Celeris streams 1581 tokens per second, the fastest output speed on BenchLeader.

What is the fastest good LLM?

Among models scoring above 55 on the BenchLeader Index, K2 Horizon 375B A23B is the fastest on 24 Oct 2026 at 982 tokens per second (index 56.3).

How is LLM output speed measured?

Output tokens per second: Artificial Analysis' measured median where it covers a model, otherwise OpenRouter's live-traffic median smoothed over the last seven days. It counts tokens streamed per second after the first one arrives, and varies by provider and time of day, so treat it as typical rather than guaranteed.

What is the difference between output speed and latency (time to first token)?

Output speed is how fast text streams once it starts, in tokens per second; latency, or time to first token, is how long you wait before the first token arrives. A reasoning model can stream quickly but still take seconds to start, because it thinks first.