GLM-4.7-Flash: VRAM requirements & speed
GLM-4.7-Flash (31.22B MoE, 3B active, mit). Native context 198k. Smallest GGUF: 18.31 GB.
Memory needed by quantization and context
GGUF file size plus f16 KV cache (and fixed overhead of ~1 GB per GPU not shown). Source: unsloth/GLM-4.7-Flash-GGUF.
| Quant | File | +8k ctx | +32k ctx | +128k ctx |
|---|---|---|---|---|
| Q4_K_M | 18.31 GB | 18.8 GB | 20.1 GB | 25.4 GB |
| Q5_K_M | 21.41 GB | 21.9 GB | 23.2 GB | 28.5 GB |
| Q6_K | 24.69 GB | 25.1 GB | 26.5 GB | 31.8 GB |
| Q8_0 | 31.84 GB | 32.3 GB | 33.6 GB | 38.9 GB |
KV cache size by context length
Computed from the model's attention layout.
| Context | f16 | q8_0 | q4_0 |
|---|---|---|---|
| 8k | 0.44 GB | 0.24 GB | 0.12 GB |
| 32k | 1.77 GB | 0.94 GB | 0.50 GB |
| 128k | 7.10 GB | 3.77 GB | 2.00 GB |
Which hardware runs GLM-4.7-Flash? · 16k context
| Hardware | Quant | Fit | Memory | Tok/s | Speed source |
|---|---|---|---|---|---|
| RTX 5090 32GB | Q6_K | fits fully | 25.6 GB | 268 | scaled from measured |
| RTX 4090 24GB | Q5_K_M | fits fully | 22.3 GB | 257 | scaled from measured |
| RTX PRO 6000 96GB | Q8_0 | fits fully | 32.7 GB | 248 | scaled from measured |
| AMD RX 7900 XTX 24GB | Q5_K_M | fits fully | 22.3 GB | 223 | estimated |
| RTX 3090 24GB | Q5_K_M | fits fully | 22.3 GB | 192 | scaled from measured |
| Mac Studio M5 Ultra 256GB | Q8_0 | fits fully | 32.7 GB | 192 | estimated |
| Mac Studio M5 Ultra 512GB (ships late Oct 2026) | Q8_0 | fits fully | 32.7 GB | 192 | estimated |
| 2x RTX 3090 (48GB total) | Q8_0 | fits fully | 32.7 GB | 165 | estimated |
| RTX 5080 16GB | Q4_K_M | MoE offload to CPU | 19.2 GB | 108 | estimated |
| RTX 5070 Ti 16GB | Q4_K_M | MoE offload to CPU | 19.2 GB | 106 | estimated |
| RTX 5060 Ti 16GB | Q4_K_M | MoE offload to CPU | 19.2 GB | 92 | estimated |
| 16GB VRAM class (RTX 4060 Ti 16GB, 32GB RAM) | Q4_K_M | MoE offload to CPU | 19.2 GB | 80 | estimated |
| Mac Studio (2025) M3 Ultra 96GB | Q8_0 | fits fully | 32.7 GB | 79 | scaled from measured |
| Mac Studio M5 Max 128GB | Q8_0 | fits fully | 32.7 GB | 74 | scaled from measured |
| MacBook Pro 14/16 M5 Max 128GB | Q8_0 | fits fully | 32.7 GB | 74 | scaled from measured |
| MacBook Pro 14/16 (2024) M4 Max 128GB | Q8_0 | fits fully | 32.7 GB | 67 | estimated |
| RTX 5070 12GB | Q4_K_M | MoE offload to CPU | 19.2 GB | 59 | estimated |
| RTX 4070 12GB | Q4_K_M | MoE offload to CPU | 19.2 GB | 58 | estimated |
| AMD Ryzen AI Max+ PRO 495 (Ryzen AI Halo) 192GB | Q8_0 | fits fully | 32.7 GB | 56 | estimated |
| MacBook Pro 14/16 M5 Pro 48GB | Q8_0 | fits fully | 32.7 GB | 55 | estimated |
| AMD Ryzen AI Max+ 395 (Strix Halo) 128GB | Q8_0 | fits fully | 32.7 GB | 54 | scaled from measured |
| NVIDIA DGX Spark 128GB | Q8_0 | fits fully | 32.7 GB | 50 | scaled from measured |
| Mac mini M6 32GB | Q4_K_M | fits fully | 19.2 GB | 47 | estimated |
| MacBook Air M5 32GB | Q4_K_M | fits fully | 19.2 GB | 44 | estimated |
| 8GB VRAM class (RTX 4060 8GB, 32GB RAM) | Q4_K_M | MoE offload to CPU | 19.2 GB | 41 | estimated |
| Mac mini (2024) M4 Pro 64GB | Q8_0 | fits fully | 32.7 GB | 41 | estimated |
| RTX 3060 12GB (32GB RAM) | Q4_K_M | MoE offload to CPU | 19.2 GB | 34 | estimated |
| Mac mini / iMac M4 16GB | - | does not fit | - | - |
Discrete GPUs are modelled with 32 GB of system RAM available for offload. Does not fit even with offload: Mac mini / iMac M4 16GB.
MLX conversions for Apple Silicon
| Repo | Bits | Size | Runs with |
|---|---|---|---|
| mlx-community/GLM-4.7-Flash-4bit | 4-bit (4bit) | 16.9 GB | mlx-lm |
| mlx-community/GLM-4.7-Flash-8bit | 8-bit (8bit) | 31.8 GB | mlx-lm |
Verified via the Hugging Face API on 2026-10-20. Compare with the GGUF sizes above (Q4_K_M 18.31 GB, Q8_0 31.84 GB). Mac hardware pages show the MLX command; there are no MLX speeds on this site yet. MLX vs GGUF.
Links
Model card · GGUF files · Open in the advisor · last verified 2026-10-20
FAQ
How much VRAM does GLM-4.7-Flash need?
The smallest quant (Q4_K_M) is 18.31 GB, so you need roughly 20 GB for an 8k context including KV cache and overhead. At 128k context the f16 KV cache alone is 7.1 GB.
Can I run GLM-4.7-Flash on a RTX 5090 32GB?
Yes: on RTX 5090 32GB it fits fully as Q6_K at about 268 tok/s (scaled from measured). 19 of 28 hardware configurations fit it fully at 16k context.
What is the maximum context for GLM-4.7-Flash?
198k tokens natively. Memory, not the model, is usually the limit: see the KV cache table above.
All models: model index · by memory: 8 GB · 12 GB · 16 GB · 24 GB · 32 GB · 48 GB · 64 GB · 96 GB · 128 GB · 192 GB · 256 GB · 512 GB. Numbers use the same fit math as the advisor; sources and methods.