Skip to content
Model Change Index
  • OverviewNext shutdowns and latest changes
  • ShutdownsEvery date, one row per platform
  • ChangesReleases, deprecations, redirects and prices
  • LeaderboardEvery model, its scores and price
  • Check my stackYour models on one timeline
Track my stack

Mistral Small 3.2

Shut downMistral AImistral/mistral-small-3-21 platform

Next shutdown
None setNo date on 1 platform
Arena Text
1,356#95 of 128 · Arena
Context window
128Ktokens
Released
Jun 20, 2025Mistral AI

Dates on every platform

1 listings · a floor is the earliest possible date
Mistral Small 3.2 on each platform: model ID, status, dates, replacement and source
PlatformModel ID and statusDeprecatedShuts downReplacement
Mistral APIMistral AI: Models Overview (opens in a new tab) · checked Oct 14, 2026mistral-small-2506Shut downMistral Small 4

Mistral's models overview still showed the status Deprecated on 14 October 2026, after the retirement date.

The mistral-small-latest alias pointed to this model before it moved to Mistral Small 4.

What breaks when you move off Mistral Small 3.2

  • Behavior

    Magistral models always returned thinking traces. Mistral Small 4 and Mistral Medium 3.5 reason through the reasoning_effort parameter: high returns a thinking chunk before the answer, none omits it.

    Mistral Small 3.2, Magistral Small 1.2 to Mistral Small 4 · Mistral API ·

Intelligence

Third-party results, credited below
Third-party benchmark scores for Mistral Small 3.2
BenchmarkPublisherScoreRankPublished
Arena TextArena1,356 ± 3#95 of 128Oct 8, 2026

Arena (arena.ai), https://arena.ai/leaderboard

Change history

ShutdownMistral Small 3.2 and Magistral Small 1.2 retire from the Mistral APIMistral API1 break

mistral-small-2506 and magistral-small-2509 retired on 31 July 2026. Mistral names Mistral Small 4 as the replacement for both.

Platform
Mistral API
Replacement
Mistral Small 4
Announced
  • BehaviorMagistral models always returned thinking traces. Mistral Small 4 and Mistral Medium 3.5 reason through the reasoning_effort parameter: high returns a thinking chunk before the answer, none omits it.

Source: Mistral AI, Models Overview (opens in a new tab); Mistral AI, Native reasoning (deprecated) (opens in a new tab); Mistral AI, Reasoning (opens in a new tab)

Built by TeqEngine. Forward-deployed AI engineering, no model to sell. Moving an agent to a new model? Ask for a Next-Model Check.