live All models chatreasoning

Phi-4 14B

Punches far above its size on maths. Pair it with retrieval, not with trivia.

$argosy up --model phi-4-14b
20ms

p50 from the nearest warm region

428tok/s

Single stream, no batching

16K

Full window, no sliding truncation

$0.09

Per million tokens served

Punches above its size on maths and structured reasoning. Weak on world knowledge, so pair it with retrieval rather than asking it what happened in 1848.

Deploymentdep_9bf31c
Parameters14B
Quantisationfp8
LicenceMIT
Home regionord1
Extra replicas5
p50 latency22ms
p99 latency55ms
Requests / s3350
Figures are illustrative — this is a demonstration site.
Last 30 minutesord1
cache hitqueue 0ms errors 0

Put the model next to the user.

One command, thirty-one regions, and an invoice that matches what you actually served.