AI Research Atlas

GLM-4-9B

Zhipu AI / Z.ai · 5 June 2024

Open 9B model with 128K chat variant (and a 1M-token variant), 26 languages, beating Llama-3-8B on MMLU and HumanEval.

Open release of the GLM-4 family's small model, with the GLM-4 technical report (arXiv 2406.12793, 2024-06-18) following. Zhipu reports 74.7 MMLU against Llama-3-8B's 66.6 and 70.1 HumanEval against 62.2.

Date
Wednesday, 5 June 2024
Lab
Zhipu AI / Z.ai
Kind
open-weights
Access
open weights (restricted license)

Figures

MeasureValueMeasured by
MMLU (GLM-4-9B vs Llama-3-8B)74.7 vs 66.6
company-reported		company
HumanEval70.1 vs 62.2
company-reported		company

Launch date from the GLM-4 repo news (2024-06-05). Governed by the GLM-4 licence, not Apache/MIT.

Sources

  1. github.com/THUDM/GLM-4
  2. huggingface.co/THUDM/glm-4-9b
  3. arxiv.org/abs/2406.12793

This record was checked against its sources on 6 October 2026. How we check

Related