AI Progress Tracker

    Qwen3 1.7b · Qwen3-1.7B

    Alibaba · 29 Apr 2025

    Ask the Advisor about this model
    26from 2
    benchmarks

    Qwen3-1.7B is Alibaba's dense, open-weight text model with 1.7 billion parameters, released under the Apache 2.0 license. It supports switchable thinking and non-thinking modes, multilingual generation, coding and tool use, with a documented 32,768-token context window.

    context: 33Ktext

    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 questions38%28 Aug 2026View
    OTIS Mock AIME 2024-25Competition mathematics8.1%28 Aug 2026View

    Weighted mean of 2 verified benchmark results: GPQA Diamond 38%, OTIS Mock AIME 2024-25 8.1%.

    Capabilities

    • Switches between thinking and non-thinking modes for reasoning-intensive tasks and faster responses.
    • Supports 119 languages and dialects.
    • Generates and explains code.
    • Supports tool calling and agent integrations through Qwen-Agent.
    • Provides downloadable weights for local deployment and customization under Apache 2.0.

    What improved

    • Introduces integrated thinking and non-thinking modes compared with Qwen2.5 instruction models.
    • Expands documented language and dialect coverage from 29 in Qwen2.5 to 119.
    • Adds thinking-budget control to balance reasoning effort against inference cost and latency.

    Known flaws & limitations

    • The documented native context window is limited to 32,768 tokens.
    • This checkpoint accepts text rather than providing native image, audio or video understanding.
    • The model card warns that greedy decoding in thinking mode can degrade performance and cause endless repetition.
    • Thinking mode requires additional generated tokens and can increase response latency.

    Risks, safety & security issues

    • Not yet confirmed by public reporting.

    Good fit for

    • Locally hosted text assistants where a compact model is preferred.
    • Multilingual drafting, rewriting and conversational applications.
    • Coding assistance with human review and executable tests.
    • Tool-calling prototypes with application-enforced permissions.

    Poor fit for

    • Native image, audio or video understanding.
    • Tasks requiring more than the documented 32,768-token context window.
    • Strictly latency-sensitive applications when thinking mode is enabled.
    • Thinking-mode deployments that require greedy decoding.
    Security & risk dashboardLatest vulnerabilities, advisories and real-world incidents for Qwen3 1.7b

    References

    Profile last reviewed 10/21/2026 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.