Doubleword

    This model was retired on 30 October 2026 and no longer accepts API requests. Recommended replacement: GLM 5.3.

    Agentic Engineering
    Thinking Mode
    FP8
    Z.ai
    Open Weights

    GLM-5.1
    Retired

    Z.ai's next-generation flagship for agentic engineering. State-of-the-art on SWE-Bench Pro, with significantly stronger coding than GLM-5 — built for repository generation, terminal tasks, and long-horizon agentic workflows.

    Total Parameters

    FP8

    Quantized

    Context Window

    198K

    Tokens

    Intelligence

    26.1

    AA Index v4.3

    License

    Open

    Open Weights

    About

    Built for Long-Horizon Agentic Work

    GLM-5.1-FP8 is Z.ai's next-generation flagship model for agentic engineering, with significantly stronger coding capabilities than GLM-5. It achieves state-of-the-art performance on SWE-Bench Pro and leads GLM-5 by a wide margin on NL2Repo and Terminal-Bench 2.0 — making it especially strong for real-world coding, repository generation, terminal tasks, and long-horizon agentic workflows.

    GLM-5.1 is designed to stay productive over extended sessions: breaking down ambiguous problems, running experiments, reading results, identifying blockers, and improving through repeated iteration. Thinking mode is enabled by default; disable with {"chat_template_kwargs": {"enable_thinking": false}}.

    SWE
    Repo
    Term
    Tools
    Plan
    Iter

    Flagship Agentic Engineering Model

    Use Cases

    Best for sustained engineering agents

    Agentic Engineering

    State-of-the-art on SWE-Bench Pro. Built for repository generation, multi-file edits, and long-horizon coding agents.

    Terminal & Tool Use

    Leads GLM-5 by a wide margin on Terminal-Bench 2.0 — strong at sustained tool calls, shell workflows, and iterative debugging.

    Long-Horizon Reasoning

    Stays productive over extended sessions: breaks down ambiguous problems, runs experiments, reads results, identifies blockers, and improves through iteration.

    NL2Repo Generation

    Top-tier performance on natural-language-to-repository benchmarks — turn specs and prompts into full working codebases.

    Benchmarks

    Frontier Coding & Agentic Performance

    Artificial Analysis Intelligence Index v4.3 scores. SOTA on SWE-Bench Pro.

    26.1

    Intelligence Index

    Better than 92% of models

    84

    GPQA Diamond

    Better than 92% of models

    98

    τ²-Bench Telecom

    Better than 95% of models

    Category Benchmark Score Description
    Reasoning GPQA Diamond 84% Graduate-level scientific reasoning
    Reasoning Humanity's Last Exam 28% Humanity's Last Exam
    Reasoning τ²-Bench Telecom 98% AI agents in dual-control scenarios
    Reasoning AA-LCR 62% Long context reasoning evaluation
    Reasoning IFBench 76% Instruction-following accuracy
    Reasoning GDPval-AA 52% Agentic performance on real-world work tasks
    Coding SciCode 44% Python for scientific computing
    Coding Terminal-Bench Hard 43% Agentic coding & terminal use
    Knowledge AA-Omniscience Accuracy 26% Proportion of correctly answered questions
    Knowledge AA-Omniscience Non-Hallucination 71% Confidently answered questions that are correct

    Metrics sourced from Artificial Analysis and Z.ai's published evaluations. Reasoning (thinking) mode enabled.