Kimi K3 tokens per second

Live benchmark data for Kimi K3 on Ollama Pro. Speed, latency, and reliability — benchmarked every ~4 hours.

What is Kimi K3?

Kimi K3 is a reasoning model served through the Ollama Pro API. This page tracks its real inference speed — tokens per second, time to first token, and 24-hour reliability — sampled every ~4 hours on the same fixed prompt and the same measurement method used for every other model, so the numbers are directly comparable across providers. See the full methodology for how each sample is taken.

Latest TPS
108.1 tok/s
24h avg TPS
72.6 tok/s
TTFT
671ms
24h reliability
100%
Last tested
1h ago

How Kimi K3 compares

By latest measured tokens per second, Kimi K3 ranks 16 of 23 available models on Ollama Pro, and 23 of 53 across all providers benchmarked here. This is a snapshot from the last site build — the live stats above and the live leaderboard show its current position.

Embed this model's live speed

Paste this line into a README or a docs page. The badge shows the current tokens per second and the age of the last sample; it shows “no recent sample” instead of an old number once the newest successful sample is stale or missing.

[![TokenDyno tokens/sec: Kimi K3](https://api.ollamatps.com/api/badge/ollama/kimi-k3.svg)](https://tokendyno.com/models/ollama/kimi-k3/)

Tokens per second over time

Time to first token over time

Reliability

Every benchmark attempt counts here — including timeouts and errors. TPS and TTFT charts above show successful runs only, so you can still see speed data while reliability is red.

Speed is measured as output tokens ÷ generation time. Each benchmark fires one streaming chat completion with a fixed ~300-token prompt, every ~4 hours. Full methodology →