Mistral Small 4

Mistral Small 4 is the efficient tier — built to do useful work at low cost and low latency, and to run on modest hardware where deployed directly.

Mistral's design philosophy is most visible at this size: extracting more from fewer parameters, which makes their small models unusually competitive against larger rivals.

Price and speed without the sharp capability cliff that usually comes with them. For classification, extraction, routing and short-form generation it is genuinely capable rather than merely adequate.

Multilingual handling is a real advantage at this size — most small models are heavily English-weighted and degrade noticeably in other languages.

It remains a small model. Complex reasoning, long context and nuanced writing are all beyond it, and it will attempt them confidently rather than declining.

Prompts need to be more specific than a larger tier requires, and output should be validated in production.

A strong default for high-volume, well-defined tasks, particularly multilingual ones. Pair it with Large 3 behind a routing decision rather than using either alone — the cheap model handling the easy majority is the pattern that keeps quality up and cost down.