AI Progress Tracker

    GLM 4.6 · glm-4.6

    Z.ai · 30 Sep 2025

    Ask the Advisor about this model
    16from 4
    benchmarks

    GLM-4.6 is an open-weight, text-focused model from Z.ai designed for reasoning, coding and tool-using agents, with a documented 200K-token context window. GLM-4.7 is a newer family release; this profile remains specific to GLM-4.6.

    context: 200Ktext$0.60 / $2.20 per 1M tokens (in / out)

    Benchmark results

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

    BenchmarkScoreSource
    Terminal-BenchLong-horizon command-line agent tasks24.5%30 Sep 2025View
    FrontierMath3.8%8 Dec 2025View
    LMArena TextHuman preference Elo from blind chat votes (not a percentage)1440 Elo22 Oct 2026View
    LMArena WebDevHuman preference Elo for building web apps (not a percentage)1341 Elo23 Oct 2026View

    Weighted mean of 2 verified benchmark results: Terminal-Bench 24.5%, FrontierMath 3.8%.

    Capabilities

    • Processes long text inputs within a documented 200K-token context window.
    • Supports code generation, debugging and repository-level software-engineering tasks.
    • Supports tool calls during reasoning for multi-step agent workflows.
    • Handles mathematical reasoning and expert-level text questions.
    • Supports writing, rewriting and instruction-following tasks.
    • Provides publicly available model weights for self-hosted deployment.

    What improved

    • Expands the context window from GLM-4.5’s 128K tokens to 200K tokens.
    • Improves coding performance and compatibility with coding-agent workflows.
    • Adds tool use during reasoning.
    • Improves tool selection and execution in agentic workflows.
    • Improves writing style and instruction following.

    Known flaws & limitations

    • Text-only model; native image, audio and video understanding are outside its documented scope.
    • The vendor reports that real-world coding performance still trails Claude Sonnet 4.
    • Published SWE-bench Verified results show that repository-level issue resolution remains incomplete.
    • Published Humanity’s Last Exam results show substantial limitations on exceptionally difficult expert questions.

    Risks, safety & security issues

    • Not yet confirmed by public reporting.

    Good fit for

    • Coding assistance and repository maintenance with tests and human review.
    • Long-document analysis and synthesis.
    • Tool-using agents operating within constrained permissions.
    • Text drafting, rewriting and structured instruction-following workflows.

    Poor fit for

    • Applications requiring native image, audio or video understanding.
    • Unreviewed deployment of generated code or autonomous repository changes.
    • High-stakes expert decisions requiring consistently correct answers.
    Security & risk dashboardLatest vulnerabilities, advisories and real-world incidents for GLM 4.6

    References

    Profile last reviewed 10/27/2026 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.