How to Deploy GLM 5.3 Flash Locally
GLM 5.3 Flash is a MoE model with 320B total parameters and 18B active per token. At Q4 it needs 202GB of VRAM to sit entirely on the GPU; with experts offloaded to system RAM it needs only 30GB of VRAM plus 174GB of RAM. The cheapest device that holds it is the Mac Studio M3 Ultra 512GB, with no price data yet, at roughly 19.3 tokens/s. In offload mode the cheapest card that runs it is the Tesla V100 32GB, at roughly 8 tokens/s.
GLM 5.3 Flash details
How much VRAM does GLM 5.3 Flash need for local deployment?
MoE: all 320B parameters must stay resident in memory, while speed is set by the 18B active per token.
| Quantization | WeightsGB | Total at 8KGB | Total at 32KGB | Total at 128KGB |
|---|---|---|---|---|
| 1-bit | 97.6GGUF | 98.6 | 166.1 | 436.1 |
| 2-bit | 108.7GGUF | 142.4 | 209.9 | 479.9 |
| 3-bit | 147.5GGUF | 176.8 | 244.3 | 514.3 |
| 4-bit | 199.7GGUF | 201.9 | 269.4 | 539.4 |
| 5-bit | 240.3GGUF | 239.4 | 306.9 | 576.9 |
| 8-bit | 341GGUF | 355.2 | 422.7 | 692.7 |
| FP16 | 641.6GGUF | 649.3 | 716.8 | 986.8 |
Total = weights + KV cache (FP16) + 1GB runtime overhead. Weights marked GGUF are measured file sizes from the repository; the rest are estimated from bytes per parameter. Contexts beyond this model's 1M limit show a dash. Each tier takes whichever file of that bit width the repository has, preferring unsloth dynamic quants, so the exact file name varies by model.
Which GPUs can run GLM 5.3 Flash?
AmazonJP
| Device | VRAMGB | Price | Est. t/s | Suggested context |
|---|---|---|---|---|
| Mac Studio M3 Ultra 512GB | 512 | —No listings | 19.3 | 32K |
Fits when weights + KV cache + runtime overhead is at or below usable memory, single card. "Offload" on MoE models means it runs with expert weights in system RAM (only attention layers and KV cache stay in VRAM); that speed assumes 70 GB/s RAM bandwidth. The context column is the largest tier that fits.
Recommended GPUs for GLM 5.3 Flash with CPU/GPU offloading
AmazonJP
Expert weights live in system RAM (needs at least 174 GB); VRAM holds only attention layers, shared experts and the KV cache, at least 30 GB at Q4. Speed is bound by RAM bandwidth (estimated at 70 GB/s) and is far slower than a full-VRAM setup.
| Device | VRAMGB | Price | RAM neededGB | Est. t/s |
|---|---|---|---|---|
 Tesla V100 32GB | 32 | $1,141 | 174 | 7.7 |
 Arc Pro B70 | 32 | $2,074 | 174 | 6.1 |
 Radeon AI PRO R9700 | 32 | $2,244 | 174 | 6.4 |
| Ryzen AI Max+ 395 128GB | 128 | $3,913 | 174 | 4.3 |
| GeForce RTX 5090 | 32 | $7,489 | 174 | 8.5 |
| DGX Spark 128GB | 128 | $8,223 | 174 | 5.2 |
| Arc Pro B65 | 32 | — | 174 | 6.1 |
| GeForce RTX 4080 SUPER 32GB | 32 | — | 174 | 7.3 |
| GeForce RTX 5090 D | 32 | — | 174 | 8.5 |
| Instinct MI100 | 32 | — | 174 | 7.6 |
| Instinct MI50 32GB | 32 | — | 174 | 7.3 |
| Radeon PRO W7800 | 32 | — | 174 | 6.2 |
| RTX PRO 4500 Blackwell | 32 | — | 174 | 7.6 |
| A100 40GB PCIe | 40 | — | 174 | 8.4 |
| CMP 170HX 40GB | 40 | — | 174 | 8.4 |
| GeForce RTX 4090 48GB | 48 | — | 174 | 7.8 |
| Radeon PRO W7900 | 48 | — | 174 | 7 |
| RTX PRO 5000 Blackwell 48GB | 48 | — | 174 | 8.2 |
| CMP 170HX 64GB | 64 | — | 174 | 8.3 |
| Instinct MI210 | 64 | — | 174 | 8 |
| RTX PRO 5000 Blackwell 72GB | 72 | — | 174 | 8.2 |
| RTX PRO 6000D | 84 | — | 174 | 8.4 |
| RTX PRO 6000 Blackwell | 96 | — | 174 | 8.5 |
| Mac Studio M4 Max 128GB | 128 | — | 174 | 6.3 |
| Mac Studio M5 Max 128GB | 128 | — | 174 | 6.5 |
| MacBook Pro M4 Max 128GB | 128 | — | 174 | 6.3 |
| MacBook Pro M5 Max 128GB | 128 | — | 174 | 6.5 |
Deploy GLM 5.3 Flash locally with Ollama, llama.cpp or vLLM
The Ollama library has no official tag for it yet.
llama.cpp pulls the GGUF straight from Hugging Face; -c sets the context length.
llama-server -hf unsloth/GLM-5.3-Flash-GGUF:Q4_K_M -c 8192
vLLM serves the original-precision weights, which needs far more memory than GGUF and usually more than one GPU.
vllm serve zai-org/GLM-5.3-Flash
File names follow that week's model-library snapshot; the definitions are in the methodology. Methodology
FAQ
How much VRAM does GLM 5.3 Flash need?
At Q4_K_M with an 8K context GLM 5.3 Flash needs about 202GB of VRAM; 244GB leaves comfortable headroom.
What is the cheapest GPU that runs GLM 5.3 Flash?
The Mac Studio M3 Ultra 512GB: 512GB of VRAM at roughly 19.3 tokens/s. We have no price data for it yet.
Can you run GLM 5.3 Flash with Ollama?
Not yet: the Ollama library has no official tag for it. You can load the GGUF with llama.cpp instead.