GLM-4.5 · released 2025-07 · 110B parameters (13.4B active per token)
Card: 106B total / 12B active (safetensors incl. 1 MTP layer).
Source configMemory for one conversation's context, on top of the weights.
| Context | FP16 | FP8 |
|---|---|---|
| 4,096 tokens | 0.72 GB | 0.36 GB |
| 8,192 tokens | 1.44 GB | 0.72 GB |
| 32,768 tokens | 5.75 GB | 2.88 GB |
| 131,072 tokens | 23.0 GB | 11.5 GB |
| Format | Bits | Weights |
|---|---|---|
| BF16 / FP16 | 16 | 205.8 GB |
| FP8 (W8A8) | 8 | 105.2 GB |
| INT8 (W8A8 / W8A16) | 8 | 105.2 GB |
| NVFP4 | 4.5 | 61.2 GB |
| INT4 AWQ / GPTQ | 4.25 | 58.1 GB |
| GGUF Q8_0 | 8.5 | 109.3 GB |
| GGUF Q6_K | 6.56 | 84.4 GB |
| GGUF Q5_K_M | 5.69 | 73.2 GB |
| GGUF Q4_K_M | 4.89 | 62.9 GB |
| GGUF Q3_K_M | 3.91 | 50.3 GB |
| GGUF Q2_K | 3.35 | 43.1 GB |
One server, 8K context, FP8 on data center GPUs and GGUF Q4_K_M on consumer and Apple hardware.
| GPU | 1 user | 32 users |
|---|---|---|
| B200 SXM (HGX B200)FP8 (W8A8) | 1× · 182 tok/s | 1× · 33.7 tok/s |
| H200 SXM 141GBFP8 (W8A8) | 1× · 121 tok/s | 2× · 38.5 tok/s |
| H100 SXM 80GBFP8 (W8A8) | 2× · 139 tok/s | 2× · 27.6 tok/s |
| Instinct MI300X 192GBFP8 (W8A8) | 1× · 132 tok/s | 1× · 22.7 tok/s |
| A100 SXM 80GBINT4 AWQ / GPTQ | 1× · 94.2 tok/s | 2× · 26.3 tok/s |
| L40S 48GBFP8 (W8A8) | 4× · 56.8 tok/s | 4× · 13.3 tok/s |
| RTX PRO 6000 Blackwell Server Edition 96GBFP8 (W8A8) | 2× · 71.8 tok/s | 2× · 13.3 tok/s |
| GeForce RTX 5090 32GBGGUF Q4_K_M | 4× · 78.9 tok/s | 4× · 29.1 tok/s |
| GeForce RTX 4090 24GBGGUF Q4_K_M | 4× · 66.1 tok/s | 8× · 23.9 tok/s |
| Apple M4 Max (40-core GPU, up to 128GB)GGUF Q4_K_M | 1× · 26.9 tok/s | 1× · 3.8 tok/s |
| DGX Spark (GB10, 128GB unified)FP8 (W8A8) | 1× · 8.2 tok/s | > 16 |