models · zai-org

GLM-5.2

text-generation model harvested from Hugging Face (trending). Catalogued at Quarantine — unverified, pending attestation.

Compare Verify

What this registry has checked

Quarantine. We hold the publisher’s own published hash for these weights and have recorded it against this entry. We have not downloaded and hashed the weights ourselves, so the hash is the publisher’s claim, not our measurement.

Method	hf-published-sha256

Re-check this yourself → Source → Full record →

What is recorded

Publisher	zai-org
Origin	China
Licence	mit
Parameters	753,329,940,480 (753.3B)
Context window	1,048,576 tokens
Modality	text->text
Architecture	glm_moe_dsa
Layers	78
Hidden size	6144
Attention	MLA (64 heads, 64 KV)
Native precision	BF16
Pulls recorded here	674
Downloads reported upstream	982,017

Reported scores

GPQA Diamond	89.5
IFBench	73.3
Humanity's Last Exam	41.1

Reported by a third party and recorded here, not re-run by us. Source: artificialanalysis.ai/api/v2.

Measured on real hardware

None of these runs are ours. Each figure is the reporter’s, linked to their own evidence, under their own trust label. The number shown is single-stream decode — one user, concurrency 1 — because aggregate throughput across many users is a different quantity and reads as far faster than anyone will see.

Reported to local-ai-registry

Hardware	Engine	Build measured	Decode, 1 stream	Label	Reported by
RTX PRO 6000 Blackwell ×4 devices	vllm	EXL3 Trellis TR3 3.0 bpw	188 tok/s @ 8,192	validated	0xsero
Measured outside 2K–32K context — decode speed depends heavily on context, so these are not comparable with the rows above
RTX PRO 6000 Blackwell ×4 devices	vllm	ModelOpt safetensors NVFP4 non-uniform REAP	77.0 tok/s	candidate	0xsero
RTX PRO 6000 Blackwell ×4 devices	vllm	ModelOpt safetensors NVFP4 REAP	60.0 tok/s	candidate	0xsero
RTX PRO 6000 Blackwell ×4 devices	vllm	safetensors MXFP8/NVFP4/NF3 hybrid	58.2 tok/s @ 180,007	candidate	0xsero
RTX PRO 6000 Blackwell ×4 devices	llama.cpp	UD-Q2_K_MXFP4	47.9 tok/s @ 524,288	candidate	localmaxxing
RTX PRO 6000 Blackwell ×4 devices	llama.cpp	UD-IQ4_XS	39.8 tok/s @ 65,536	candidate	localmaxxing
NVIDIA DGX Spark GB10	vllm	Int4-Int8Mix	38.6 tok/s @ 200,000	candidate	localmaxxing
GeForce RTX 3090 ×10 devices	llama.cpp	Q3_K_M	22.1 tok/s @ 262,144	candidate	localmaxxing
NVIDIA DGX Spark GB10 ×3 devices	vllm	hybrid ModelOpt+AQLM NVFP4 hot experts + AQLM 2-bit cold experts	21.0 tok/s	candidate	mialabs

From local-ai-registry (MIT), commit 124959d: 79 runs across 6 machines, 1 validated — model revision and runtime pinned, launch accepted — and 78 candidate, their label for useful evidence without a reproducible-launch promise.

Elsewhere in the registry

All models → Browse with filters → What will run on my hardware? →

Registry record as of 2026-09-01. Identity, licence and parameter count are properties of a release and do not move; tier can, and the verify link above re-checks it against the live registry rather than this page.