Mistral AI · Small and fast

Mistral Small 4 API pricing and specs

An Apache-licensed mixture-of-experts model with 6.5B active parameters that switches between instruct and reasoning modes.

Input

⁨$0.15⁩

Cached input

—

Output

⁨$0.60⁩

US dollars per million tokens, standard tier, short-prompt rate

It replaces Mistral Small 3.2 and Magistral Small. Cached-input pricing is not published per model.

Spec sheet

API model string: mistral-small-2603
Provider: Mistral AI
Context: 256K tokens
Max output: —
Released: March 16, 2026
Batch API: 50% off, results within hours
Documented agent features: Tool callingStructured outputAdjustable reasoning

What a task costs

Four reference workloads priced at this model's list rates, with caching applied after the first call. Your numbers will differ; the calculator lets you change every assumption.

Workload	Per task	Per 1,000 tasks
Support bot 4 calls, short tool lookups	⁨$0.006⁩	⁨$6⁩
Coding agent 30 calls, files and test output	⁨$0.526⁩	⁨$526⁩
Browser agent 25 calls, screenshots and page text	⁨$0.216⁩	⁨$216⁩
Research agent 15 calls, long search results	⁨$0.162⁩	⁨$162⁩