Magistral Small 1.2

Magistral Small 1.2 is a reasoning-oriented language model released by Mistral as part of its Magistral line. Within that family, the "Small" designation marks it as the compact tier, sitting below Mistral's larger Magistral models in parameter scale and intended for deployments where response latency and serving cost matter more than maximum capability. The "1.2" suffix indicates an iteration on earlier Magistral Small releases rather than a new family. Magistral models are built around explicit chain-of-thought style reasoning, where the model produces intermediate reasoning steps before a final answer. In measured serving performance, Artificial Analysis records Magistral Small 1.2 at roughly 141 output tokens per second with a time to first token of about 709 ms. That throughput figure is relevant for a reasoning model, since reasoning traces consume output tokens — faster generation directly reduces the wall-clock cost of longer deliberation. Other attributes such as context window length and modality support are not tracked in our database for this entry, so check the provider documentation linked from the pricing table for those details. In practice, small reasoning models like Magistral Small 1.2 are used where a task benefits from step-by-step derivation — math word problems, multi-constraint logic, structured data extraction with validation rules — but where routing every request to a large reasoning model would be impractical. Compared with Mistral's larger Magistral tier, this model trades headroom on hard problems for lower latency and cheaper generation; compared with Mistral's non-reasoning small models, it spends more tokens per request in exchange for better performance on problems that require intermediate work.