AI Progress Tracker

    Qwen3 30b A3b Instruct · Qwen3-30B-A3B-Instruct-2507

    Alibaba · 29 Jul 2025

    Ask the Advisor about this model
    63from 6
    benchmarks

    Qwen3-30B-A3B-Instruct-2507 is Alibaba's text-only mixture-of-experts instruction model, with approximately 30 billion total parameters and 3 billion activated per token. It operates exclusively in non-thinking mode and supports a native context window of 262,144 tokens.

    context: 262Ktext

    Benchmark results

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

    BenchmarkScoreSource
    AIMECompetition mathematics61.3%Jul 2025View
    GPQA DiamondPhD-level science questions70.4%Jul 2025View
    LiveCodeBenchFresh competitive-programming problems43.2%Jul 2025View
    MMLU-ProBroad expert knowledge and reasoning78.4%Jul 2025View
    OTIS Mock AIME 2024-25Competition mathematics62.2%30 Aug 2026View
    LMArena TextHuman preference Elo from blind chat votes (not a percentage)1384 Elo22 Oct 2026View

    Weighted mean of 5 verified benchmark results: AIME 61.3%, GPQA Diamond 70.4%, LiveCodeBench 43.2%, MMLU-Pro 78.4%, OTIS Mock AIME 2024-25 62.2%.

    Capabilities

    • Instruction-following and open-ended text generation in non-thinking mode.
    • Multilingual knowledge and text understanding.
    • Mathematical, scientific and coding problem-solving.
    • Long-document processing with native 262,144-token context support.
    • Tool-use workflows through compatible external integrations.

    What improved

    • The vendor reports stronger instruction-following, reasoning, mathematics, science, coding and tool use than the preceding Qwen3 release.
    • Expanded long-tail knowledge coverage across multiple languages.
    • Improved alignment with user preferences in subjective and open-ended tasks.
    • Enhanced long-context understanding with native 256K context support.

    Known flaws & limitations

    • Supports only non-thinking operation; it does not provide the original model's switchable thinking mode.
    • Text-only architecture does not natively process images or audio.
    • Tool execution requires an external integration rather than the model weights alone.
    • Full-length 262,144-token deployment requires sufficient serving resources; deployment guidance recommends reducing context length when memory is constrained.

    Risks, safety & security issues

    • Not yet confirmed by public reporting.

    Good fit for

    • Multilingual assistants and instruction-driven text applications.
    • Coding assistance and mathematical or scientific question answering with independent validation.
    • Long-document summarization and text analysis.
    • Self-hosted tool-using assistants with an external execution layer.

    Poor fit for

    • Applications requiring native image, audio or video understanding.
    • Workflows requiring a switchable thinking mode.
    • Full-context deployments with insufficient serving memory.
    Security & risk dashboardLatest vulnerabilities, advisories and real-world incidents for Qwen3 30b A3b Instruct

    References

    Profile last reviewed 10/24/2026 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.