Skip to content
NEWLive updates feed from Hugging Face, arXiv and GitHub  See what changed →

zai-org · glm4_moe_lite

GLM-4.7-Flash

GLM-4.7-Flash by zai-org: a 31.2B-parameter mixture-of-experts open-weight model under the mit license.

mitMixture of experts203K contextAuto-imported · not yet reviewed

1,861,863 downloads · 1,844 likes · Model card · synced 2026-09-19

Specifications

Parameters
31.2B (2B active)
Architecture
moe · glm4_moe_lite
Layers
47
Hidden size
2,048
KV heads · head dim
20 · 102
Vocabulary
154,880
Native dtype
Context window
202,752 tokens
Released
2026-01-19
Last modified on Hub
2026-01-29

Features & licensing

License
mit
Commercial use
Yes
Modalities
text
Tool / function calling
No
Reasoning mode
No
Languages
EN, ZH
Quantised variants on Hub
AWQ, FP8, GGUF, bitsandbytes
Pipeline
text-generation
transformerssafetensorsglm4_moe_litetext-generationconversationalenzheval-resultsendpoints_compatible

Task fit

Editorial scores (0–100) used by the recommendation engine.

  • Chat70
  • RAG / Q&A70
  • Code70
  • Summarisation70
  • Extraction70
  • Agents70

Hardware to run GLM-4.7-Flash

Weights need about 69 GB at FP16, 36 GB at INT8 and 21 GB at INT4. The KV cache adds roughly 384 MB per 1,000 tokens per request. For a reference workload of 5 requests per second with 1,500 input and 300 output tokens, total GPU memory is around 38 GB, which fits on 1 × NVIDIA A100 (80 GB).

GPU configurations at the reference workload
ConfigurationVRAMUtilisationEst. first tokenCloud / month
2 × NVIDIA L4 (24 GB)48 GB79 %~0.5 s$1,168
1 × NVIDIA L40S (48 GB)48 GB79 %~0.2 s$1,387
1 × NVIDIA A100 (80 GB)recommended80 GB47 %~0.1 s$2,336
1 × NVIDIA H100 (80 GB)80 GB47 %~0.0 s$3,285

Estimates only. Run an assessment for your own traffic.

Auto-imported from Hugging Face. Architecture, license and feature data are synced from the model repository; the task-fit scores below are provisional defaults until a curator reviews them.