This page checks the Xiaomi MiMo API periodically and reports whether it is answering right now, alongside latency, throughput and answer correctness measured from a single European egress.
P50 per bucket. Failed runs are excluded — an outage is counted as availability, not as latency. Lower is better.
Output tokens per second over the decode window. Higher is better.
P50 end-to-end, request sent to last token. Most of the wait is getting to the first token; the gap between this plot and the time-to-first-token plot is what decoding adds. Output caps at 150 tokens, so check a step change here against the throughput plot before calling it a slowdown. Failed runs are excluded. Lower is better.
Time to complete the TCP handshake on port 443 — no TLS, no HTTP, no auth, no tokens. Each Xiaomi MiMo edge is paired with an independent reference host in the same city, so a route problem, or an outage on our side, shows up as its own problem and not as MiMo's. Only Singapore serves the inference this page measures; Amsterdam is the same service from another region, for comparison. Lower is better.
Time to first token, split into the measured TCP handshake to MiMo's edge and the remainder. Both halves come from the same 5-minute cycle.
Every probe this page sends, priced from the usage MiMo reported on it — both models.
An independent project. Not operated by, endorsed by or connected with Xiaomi. “Xiaomi” and “MiMo” are trademarks of their respective owner, named here only to identify what is measured.