DOOMBENCH.COM
Z.ai 🇨🇳 · GLM

GLM-5.2

Z.ai's open-weight flagship for million-token long-horizon coding, tool use, and locally deployable agent workloads.

DOOM SCORE84.3out of 100model risk profile, not the overall index
0 comments · 0 votesOpen discussion

Public discussion is readable by everyone. Sign in to comment, reply, or vote.

No comments yet. Start the discussion.

Send detailed or private feedback →
CURRENT ASSESSMENT · REVISION 4

Why this model scores 84.3

GLM-5.2 has strong long-context coding and tool-use evaluations, a stable million-token context, public weights, MIT licensing, and support across local inference frameworks. Open redistribution and offline deployment materially increase diffusion and make provider-level monitoring or withdrawal difficult, while benchmark-only evidence limits confidence in the highest real-world capability claims.

Capability84
Autonomy87
Deployment86
Misuse potential74
Control difficulty88
WATCH AND EXPLORE

Reviewed videos from Artificial Intelligence Videos. These links add context and never change a DoomBench score.

Thumbnail for AI Just Entered A New Era
7:27

AI Just Entered A New Era

Károly Zsolnai-Fehér examines GLM 5.2, its training methods, open weights, hardware requirements and limits of model comparisons.

The full reviewed video examines the exact GLM 5.2 release, long-task training, open-weight deployment hardware and the limitations of the presenter’s own comparisons.
Thumbnail for GLM 5.2 vs Opus 4.8 Across Real Workloads
37:14

GLM 5.2 vs Opus 4.8 Across Real Workloads

Pat Simmons blind-tests GLM 5.2 and Opus 4.8 on coding builds and knowledge work, then compares quality, failures, token use and cost.

Directly tests this exact GLM-5.2 version across coding and knowledge-work tasks, retaining one-shot, task-selection and individual-failure limitations.
Thumbnail for GLM-5.2: Open Weights, Near-Frontier Intelligence - Zixuan Li
13:46

GLM-5.2: Open Weights, Near-Frontier Intelligence - Zixuan Li

Zixuan Li outlines GLM-5.2 capabilities, open-weight deployment and fine-tuning rationale, plus a coding harness supporting multiple models.

Zixuan Li directly presents GLM-5.2 reasoning, agentic tasks, open-weight deployment and fine-tuning tradeoffs, explaining the exact assessed version; keynote capability claims are not independently validated.
Thumbnail for What It Takes to Run Frontier AI at Home
1:35:53

What It Takes to Run Frontier AI at Home

0xSero and David Ondrej compare local AI hardware, model compression, inference speed, cost, power and upgrade paths for self-hosting.

The reviewed video explicitly discusses compressing GLM 5.2 for local inference and explains the memory, bandwidth and hardware constraints of running this exact open model.
Thumbnail for How GLM 5.2 Changes the Local AI Cost Equation
22:45

How GLM 5.2 Changes the Local AI Cost Equation

Greg Isenberg and Amir Motahari test GLM 5.2, explain Cursor and Codex setup, compare token costs and show a practical model-chaining workflow.

Isenberg and Motahari directly test GLM 5.2, explain access through coding tools and model chaining, and discuss token costs and the exact model's missing vision capability.
Thumbnail for Anthropic Sonnet 5: Agentic Coding, Pricing and Trade-offs
10:04

Anthropic Sonnet 5: Agentic Coding, Pricing and Trade-offs

TheAIGRID discusses claimed Sonnet 5 improvements, agentic task costs, model comparisons and cybersecurity safeguards.

The review explicitly compares open-weight GLM 5.2 as a coding alternative and discusses safeguards, explaining the assessed version's access and control tradeoffs.
MODEL-ATTRIBUTED EVIDENCE

News tied to GLM-5.2

The model score of 84.3 rates this model's risk profile. The overall Doom Index of 66.8 measures the complete temporally weighted evidence record. These values answer different questions.

NET MODEL-ATTRIBUTED INDEX CONTRIBUTION+0.24

Each article's current Doom Index contribution is divided equally among the exact models named on that article. This prevents multi-model evidence from being claimed in full on several model pages. Model risk scores use a bounded temporal offset around their technical profile, but never feed back into the overall index.

AUDIT TRAIL

Model score history

  1. R4
    Doom Score 84.3

    Exact-version evidence chronology replayed after run doombench-hourly-news-20260812-142416 under temporal-monthly-pressure-v4.

    12 Aug 2026
  2. R3
    Doom Score 84.1

    Exact-version evidence chronology replayed after run doombench-hourly-news-20260812-132112 under fixed-sensitivity-v3.

    12 Aug 2026
  3. R2
    Doom Score 83.9

    Full-corpus evidence recalculated after run intensive-backfill-20260811-191225 under bounded-corpus-v2.

    11 Aug 2026
  4. R1
    Doom Score 84.2

    Initial exact-version profile from Z.ai's dated release, public weights, MIT license, and deployment documentation.

    11 Aug 2026

Share this page

DoomBench social sharing card for GLM-5.2.

  1. GLM-5.2 by Z.ai has a DoomBench model risk score of 84.3 out of 100, based on five transparent version-specific dimensions rather than the overall index.

  2. GLM-5.2's highest current DoomBench dimension is control difficulty at 88.0 out of 100; the profile publishes every component score and its editorial rationale.

  3. DoomBench links 2 source-backed evidence items to GLM-5.2, while keeping the model's risk profile separate from each item's contribution to the live Doom Index.

    https://www.doombench.com/models/z-ai-glm-5-2