Skip to content
Trending storyDeveloping

Epoch AI ranks Zhipu's GLM-5.3 top open-weight model on its Cyber ECI

2 articles2 sourcessince Oct 15Last article 1h ago ·

Overview

AISummary of 2 articles

Epoch AI says Z.ai's GLM-5.3 is the highest-scoring open-weight model on its cyber-specific Epoch Capabilities Index, with a score of 160, above OpenAI's GPT-5.4, released in March.

Epoch estimates that open-weight models trail the closed-weight frontier in cyber capabilities by 6–7 months.

The 6–7 month estimate comes from cybersecurity benchmark results alone, using a joint fit of the ECI on general and cyber benchmarks, and is on the same scale as the general ECI, Epoch says in a separate report. The GLM-5.3 ranking appears in a single Epoch post on X, which gives no further method details.

Written by AI from the articles below · updated Oct 16, 3:28 PM ET

Check the sources:

Latest developmentOct 16, 2:39 PM ET
Epoch AI says open-weight models trail the closed frontier in cyber by 6–7 months

Developments

2 developments

  1. Oct 16, 2:39 PM ET · 1 article
    Epoch AI reports Zhipu's GLM-5.3 tops open-weight models on its Cyber ECI
    Epoch AI: Epoch AI says open-weight models trail the closed frontier in cyber by 6–7 months
  2. Oct 15, 8:00 PM ET · 1 article
    Open-weight models trail closed models in cyber capabilities by 6–7 months
    Epoch AI: Open-weight models trail closed models in cyber capabilities by 6–7 months

Article timeline

The articles in this story. Times are ET.

Oct 16
  1. Epoch AIOfficial
    Epoch AI says open-weight models trail the closed frontier in cyber by 6–7 months

    AIEpoch AI estimates open-weight models lag the closed frontier by about 6–7 months on cyber capabilities. Z.ai's GLM-5.3 is the highest-scoring open-weight model on Epoch's Cyber ECI, with a score of 160. That score is higher than OpenAI's GPT-5.4, released in March.

    Image from @EpochAIResearch's post
Oct 15
  1. Epoch AIOfficial
    Open-weight models trail closed models in cyber capabilities by 6–7 months

    AIEpoch AI estimates that open-weight models trail closed-weight models by 6–7 months on cybersecurity capabilities, measured with a cyber-specific version of its Epoch Capabilities Index (ECI). The estimates come from cybersecurity benchmark results alone, using benchmark parameters from a joint fit of the ECI on general and cyber benchmarks. Scores are on the same scale as the general ECI.

Heat trend

Not enough continuous observations to show a trend yet.