AI Progress Tracker

    Mistral Small 3.2 · Mistral-Small-3.2-24B-Instruct-2506

    Mistral · 20 Jun 2025

    Ask the Advisor about this model
    42from 2
    benchmarks

    Mistral Small 3.2 is a 24-billion-parameter, open-weight instruction model supporting text and image inputs, multilingual interaction, and function calling. It refines Mistral Small 3.1 with better instruction following, fewer repetitive generations, and a more robust function-calling template.

    context: 128Ktext

    Benchmark results

    Every figure links to where it was published. Compare like with like — harnesses and prompt setups differ between vendors.

    BenchmarkScoreSource
    GPQA DiamondPhD-level science questions49.1%28 Aug 2026View
    OTIS Mock AIME 2024-25Competition mathematics30.3%30 Aug 2026View

    Weighted mean of 2 verified benchmark results: GPQA Diamond 49.1%, OTIS Mock AIME 2024-25 30.3%.

    Capabilities

    • Processes text and image inputs to produce text responses.
    • Supports multilingual conversation and instruction following.
    • Supports function calling for integration with external tools.
    • Supports structured output generation.
    • Provides downloadable weights under the Apache 2.0 license for self-hosting and customization.

    What improved

    • Improves instruction following relative to Mistral Small 3.1.
    • Reduces repetition errors and infinite-generation behavior.
    • Introduces a more robust function-calling template.

    Known flaws & limitations

    • Repetition errors are reduced rather than eliminated.
    • Instruction following remains imperfect in the published evaluations.
    • Function-calling improvements do not guarantee correct tool selection or arguments.

    Risks, safety & security issues

    • Not yet confirmed by public reporting.

    Good fit for

    • Multilingual conversational assistants.
    • Image-grounded question answering and document assistance.
    • Tool-enabled workflows with application-side validation.
    • Self-hosted applications requiring customizable, open-weight models.

    Poor fit for

    • Unsupervised high-stakes decisions requiring guaranteed correctness.
    • Privileged tool execution without argument validation and authorization checks.
    • Generation workflows without repetition detection or output limits.
    Security & risk dashboardLatest vulnerabilities, advisories and real-world incidents for Mistral Small 3.2

    References

    Profile last reviewed 10/20/2026 and re-checked automatically every few hours against the vendor's published documentation, papers and public leaderboards. Scores come from the cited benchmark results above — confirm limits and pricing against the provider's own documentation before building on them.