GLM-5.3

What this model is

The official GLM-5 technical report states 744B total parameters and 40B active parameters. The 78-layer sparse model uses 256 routed experts with eight selected per token and a 1M-token context.

The architecture is mixture of experts. Memory must hold the full weight set, while active parameters affect the speed heuristic. Experts can be offloaded to system memory, but that trades speed for capacity.

How this GLM model actually runs locally

GLM-5.3 is the full 744B-total model, not another Flash size. Its 228 GB lowest-bit GGUF is larger than the usable memory of nearly every consumer GPU and every 96-256 GB compact system. A 512 GB Mac Studio or a multi-GPU server is the realistic local floor. The calculated 256 GB pairing is intentionally shown as a failure instead of implying that the model may squeeze in.

The official card supports SGLang, vLLM, Transformers, KTransformers, and Unsloth, and llama.cpp now has the required architecture path. Ollama remains cloud-only. A 418 GB community MLX 4-bit conversion is explicitly aimed at a 512 GB Mac Studio, but this calculator keeps the larger GGUF package sizes so the memory answer does not understate the download.

Runtime support was verified on 2026-11-02. Exact package names and kernel support can change quickly; use the source links below for the model, quantization, and runtime projects.

Memory and quantization

Decimal GB from Unsloth's sharded GGUF repository on 2026-11-02. Even the lowest listed dynamic 1-bit-class package is 228 GB before KV cache and runtime overhead.

Estimate formula: published weight files + 0.7361 GB per 8K tokens × context / 8K + 1.5 GB runtime overhead. Sizes are decimal GB, not GiB. At 8K context, the smallest listed build is about 230.7 GB before OS and application reserve.

Conservative FP16 estimate from the 78-layer MLA configuration: 512 compressed KV dimensions plus 64 RoPE dimensions per token. It excludes the MTP layer and runtime workspace.

Ollama tags were checked against the linked Ollama library pages on 2026-11-02. LM Studio's lms get suffix selects a file when the search results expose that exact quantization; confirm the local key with lms ls.

KV method source: https://huggingface.co/zai-org/GLM-5.3/raw/main/config.json

Speed heuristic

The model page does not have one universal speed number: memory capacity and bandwidth belong to the machine, not the checkpoint. Open a model-and-hardware page for the bandwidth-based range and its conditions.

No laptop-scale measurement is claimed. The 40B active path can generate quickly when experts and attention kernels stay resident, but the 744B stored set forces a 512 GB-class unified-memory machine or a multi-GPU server.

Active path: 40B parameters. This is a relative speed signal only; it is not a tokens/sec benchmark. No verified measurement is substituted for an estimate.

Public speed checks for this model

No verified measurement yet for this exact model record.

If the full weights do not fit in accelerator memory but do fit in system memory, the calculator marks the result as a tight CPU-offload fit. Expect substantially slower generation and verify the runtime's MoE offload support.

Runtimes

• llama.cpp
• vLLM
• SGLang

No local Ollama tag. Ollama lists GLM-5.3 as cloud-only. A 512 GB Mac can use a low-bit GGUF through llama.cpp; server deployments can use vLLM or SGLang.

License restrictions

The custom GLM-5.3 license permits commercial use, modification, and redistribution. A Model-as-a-Service operator whose organization and affiliates exceed US$10 billion revenue over 12 consecutive months must pass Z.AI's security review before commercial use.

This is a plain-language summary, not legal advice. Read the linked license before commercial deployment or redistribution.

Known issues

• This is not a consumer-GPU model. The smallest listed GGUF is about 228 GB, so 256 GB machines do not have enough headroom and ordinary 24-96 GB systems are far below the floor.
• A published 4-bit MLX conversion is about 418 GB and its model card targets a 512 GB Mac Studio. The calculator keeps the GGUF sizes so it does not understate the download.
• Ollama's glm-5.3 library entry is cloud-only. Local use requires llama.cpp or a server runtime.
• The custom license is permissive for most uses but adds a security-review trigger for very large Model-as-a-Service operators.

If it does not fit, try these smaller records

These alternatives have a lower recommended memory target in the current catalog. Open each page to compare the quantization-specific fit before downloading.

FAQ

Model-specific questions

Sources

• https://huggingface.co/zai-org/GLM-5.3
• https://huggingface.co/zai-org/GLM-5.3/raw/main/LICENSE
• https://arxiv.org/abs/2602.15763
• https://huggingface.co/unsloth/GLM-5.3-GGUF
• https://huggingface.co/mlx-community/GLM-5.3-4bit
• https://ollama.com/library/glm-5.3
• https://huggingface.co/zai-org/GLM-5.3/raw/main/config.json

This page is an independent compatibility guide, not official documentation. Verify the primary source before downloading or deploying a model.

Check a specific machine