Skip to content
Llama Builds

Z.ai · Released August 2026

GLM-5.3-Flash

What it takes to run GLM-5.3-Flash locally: memory by quantization, the GPUs that handle it, and how fast it runs.

Updated

Parameters
320B (18B active)
Architecture
Mixture of experts
Context
1.0M tokens
License
MIT license
Memory at Q4_K_M
~201GB

What do you need to run GLM-5.3-Flash?

GLM-5.3-Flash needs about 201GB of VRAM at Q4_K_M with 8K context (194GB for the weights plus KV cache and overhead). Because it is a mixture-of-experts model with 18B active parameters, it generates text much faster than a dense model of the same size, and it tolerates partial CPU offload better.

See the full hardware guide for GLM-5.3-Flash

What to buy

GPUs for GLM-5.3-Flash

Unified memory

M3 Ultra 512GB

512GB unified · $9,499 MSRP

Quant
Q6_K
Speed (est.)
~17 tok/s
Runs well
Find it used on eBay

Multi-GPU

8× RTX 5090

256GB total · ~$34,400

Quant
Q4_K_M
Speed (est.)
~29 tok/s
Runs well
Check price at Amazon

Speeds are estimates at the quantization shown with 8K context. We may earn a commission when you buy through these links. Learn more.

Memory

GLM-5.3-Flash VRAM requirements by quantization

Lower-bit quantizations shrink the model at some cost to quality. Q4_K_M is the usual sweet spot. Totals include the KV cache for the context length shown plus runtime overhead.

Memory needed for GLM-5.3-Flash at each quantization level
QuantizationQualityWeightsTotal @ 8K ctxTotal @ 32K ctx
Q2_KRough120 GB124 GB125 GB
Q3_K_MUsable156 GB161 GB162 GB
Q4_K_MRecommendedDefault194 GB201 GB201 GB
Q5_K_MVery good228 GB236 GB236 GB
Q6_KNear-lossless264 GB273 GB273 GB
Q8_0Practically lossless340 GB351 GB351 GB
BF16Full precision640 GB660 GB660 GB

KV cache note: Hybrid: 34 of 45 layers use Kimi Delta Attention with a fixed-size state, and only 11 layers use sparse MLA with a growing cache. Assuming a 512-value compressed latent per MLA layer as in GLM-5, with no RoPE part because these layers use NoPE: 11 x 512 x 2 bytes = 11,264 bytes per token, about 11.8GB at 1M tokens. Not yet confirmed against a runtime log.

Performance

How fast GLM-5.3-Flash runs

Single-user generation speed on popular GPUs that fit the model entirely in memory, at the best quantization each can hold.

  1. M5 Ultra 256GB

    Q3_K_M · 256GB

    ~39tok/s

  2. M5 Ultra 512GB

    Q6_K · 512GB

    ~25tok/s

  3. M3 Ultra 512GB

    Q6_K · 512GB

    ~17tok/s

Estimated (method)

Overview

About GLM-5.3-Flash

GLM-5.3-Flash is Z.ai's August 2026 mid-size model and its first natively multimodal one: 320B total parameters with 18B active, a 1M-token context, a thinking mode, and an MIT license. It ranks just behind the largest open models on the Artificial Analysis Intelligence Index while needing well under half their memory. Most of its layers use linear attention with a fixed-size state, so long contexts are cheap. Expect about 200GB at Q4_K_M with 8K context, which means a 256GB Mac Studio or a multi-GPU server; the smallest 1-bit GGUF is 93GB and fits 128GB machines, at a clear quality cost.

Strengths

  • +Near-frontier coding and agentic performance with 18B active parameters
  • +Image input and a 1M-token context
  • +Small KV cache: only 11 of 45 layers keep a per-token cache
  • +MIT license

Weaknesses

  • −About 200GB at Q4_K_M, beyond all but the largest workstations and Macs
  • −The official FP8 weights are far too large for home use, so local setups rely on community GGUF quants
  • −llama.cpp support for its hybrid attention is new
CodingAgents & toolsReasoningVisionLong contextChatImage inputToggleable reasoning

Setup

How to run GLM-5.3-Flash

llama.cpp
$ llama-server -hf unsloth/GLM-5.3-Flash-GGUF:Q4_K_M -c 8192 -ngl 99

Weights: zai-org/GLM-5.3-Flash · GGUF: unsloth/GLM-5.3-Flash-GGUF

Alternatives

Models like GLM-5.3-Flash

DeepSeek

DeepSeek V4 Flash

284B (13B active) · 1.0M ctx

~168GBat FP4
ReasoningCodingAgents & tools

Z.ai

GLM-5.3

744B (40B active) · 1.0M ctx

~466GBat Q4_K_M
CodingAgents & toolsReasoning

Alibaba Qwen

Qwen3.5 122B-A10B

122B (10B active) · 256K ctx

~77GBat Q4_K_M
ChatReasoningCoding

Z.ai

GLM-4.5-Air

106B (12B active) · 128K ctx

~68GBat Q4_K_M
CodingAgents & toolsReasoning

Frequently asked questions

How much VRAM do you need to run GLM-5.3-Flash?

About 201GB at Q4_K_M with 8K tokens of context: 194.0GB of weights, 0.1GB of KV cache, and 6.4GB of runtime overhead. Higher-precision quants and longer contexts need more; see the table above.

Is GLM-5.3-Flash free for commercial use?

GLM-5.3-Flash is released under the MIT license. That is a permissive license, so commercial use is allowed.

How long a context does GLM-5.3-Flash support?

Up to 1,048,576 tokens (1.0M). Long contexts need a lot of extra memory for the KV cache, so most local setups run 8K to 32K.

Sources: GLM-5.3-Flash model card (Hugging Face), Unsloth: How to run GLM-5.3-Flash locally, runaihome: GLM-5.3-Flash local hardware guide. Capability score basis: Artificial Analysis Intelligence Index v4.3: 42 (September 2026), just behind GLM-5.3 and Kimi K3 at 44, mapped onto the Llama Builds scale.