Main area
Agency
AI Models Mistral

Magistral — Mistral's reasoning line explained

Martin Rau ·

Magistral isn’t a single version — it’s a line: Mistral AI’s family of models built for reasoning. This page describes the line as a whole: what it’s designed for, what sets it apart from other reasoning models, and when you’d pick it. For version-specific details on the current model, see the glossary entry → Magistral.

What Magistral is built for

Reasoning models produce an explicit chain of thought before the actual answer. Magistral’s emphasis is on keeping that chain transparent and traceable — you see how the model reaches a conclusion, rather than trusting a black box. That matters in practice the moment you have to check results or justify them to someone else.

The second, rarer building block is multilingualism. Magistral is deliberately built for reasoning in European languages, where many competitors optimise primarily for English. Combined with Mistral’s base in Paris and its GDPR obligations, that gives it a profile that can be a real selection criterion for European applications — not just a compliance checkbox.

Suitability profile

Suitability profile: Magistral
CodingReasoningTextVisionSpeedKosten-Eff.
  • Coding 4 / 5 · Claude Opus 5
  • Reasoning 4.5 / 5 · Claude Opus 5
  • Text 4 / 4.5 · Claude Opus 5
  • Vision 2 / 4 · Claude Opus 5
  • Speed 3 / 4 · Codestral
  • Kosten-Eff. 3.5 / 4.5 · DeepSeek V4-Pro

Eignung 0–5 · redaktionelle Einordnung, kein Benchmark · gestrichelt = Feld-Bestwert je Achse

The axes are an editorial read, not a benchmark. Magistral is balanced with a clear tilt toward reasoning and text; speed and coding sit below the coding specialists, and vision isn’t the focus. A thinking tool, not autocomplete.

Positioning in the field

Magistral against code and frontier models
Frontier Allrounder Volumen Geschwindigkeit / Kosten-Effizienz → Fähigkeit / Reasoning ↑ Claude Opus 5.5 DeepSeek V4-Pro Magistral Codestral

Redaktionelle Einordnung, kein Benchmark

Magistral sits in the all-rounder band, not quite at the frontier edge of the most expensive flagships. That’s the honest read: at the very top tier, Mistral doesn’t quite reach Claude Opus on raw reasoning. What you get in return is the open Small variant for self-hosting and the European data posture — criteria a single benchmark number won’t capture.

Picking a model: when Magistral, when something else

  • Magistral, when the task calls for multi-step reasoning — maths, logic, structured analysis and planning — and you need a traceable chain of thought.
  • Magistral Small, when you work with sensitive data or in European languages and want to self-host or fine-tune the model.
  • Codestral, when it’s about code completion in the editor. Codestral is fast and tailored to code; Magistral prefers to think things through.
  • Claude Opus or another frontier model, when you need the absolute top reasoning performance and data sovereignty or self-hosting don’t factor in.
  • For a broader read on all the families, see the overview of AI model families.

FAQ

What exactly is Magistral?
Magistral is Mistral AI's reasoning line: models that produce an explicit chain of thought before answering. First shipped June 2025 in Small (open) and Medium (API-only).
What sets Magistral apart from Codestral?
Codestral is a code specialist without explicit reasoning. Magistral is built for multi-step inference — even on code, structured thinking rather than fast autocomplete.
Is Magistral open source?
Partly: Magistral Small (24B) is Apache 2.0 and self-hostable. Magistral Medium is the stronger enterprise variant, API-only.
Why does European data sovereignty matter for Magistral?
Mistral AI is an EU company (subject to GDPR), plus open self-hostable Small weights — a genuine selection criterion against US providers.
Did Magistral get absorbed into Mistral Small 4?
Not entirely: Small 4 (March 2026) folds Magistral reasoning in as a switch inside a general-purpose model; Magistral remains a line of its own.

Topic overview

AI ModelsMistral