community pool / model answer
GLM-5.3-Flash
chat · zai-org/GLM-5.3-Flash
122.7 tok/s · model median · 37 runs in the model record.
| Reference rig | Quantization | Median result | Basis | TTFT (ms) | Peak VRAM | Max context | Engines |
|---|---|---|---|---|---|---|---|
| GB10 Grace Blackwell 128GB | 4-bit | 23.5 tok/s | reported | - | 99.1 GB | 262,144 | vllm |
| GB10 Grace Blackwell 256GB | 3-bit | 32.6 tok/s | measured | 419 | 105 GB | 1,000,000 | vllm |
| GB10 Grace Blackwell 512GB | 3-bit | 79.7 tok/s | measured | 664 | - GB | 300,000 | vllm |
| M3 Ultra 512GB | 4-bit | 32.9 tok/s | reported | 1,667 | - GB | 2,048 | mlx |
| Radeon AI Pro R9700 32GB ×3 | 2-bit | 25.7 tok/s | measured | 3,930 | 87.4 GB | 131,072 | llama.cpp |
| RTX PRO 6000 Blackwell 96GB ×2 | 3-bit | 149.7 tok/s | measured | 191 | 186.9 GB | 700,000 | vllm |
| RTX PRO 6000 Blackwell 96GB ×2 | 4-bit | 443.1 tok/s | measured | 125 | 186.9 GB | 8,192 | sglang, vllm |
| RTX PRO 6000 Blackwell 96GB ×4 | 8-bit | 146.8 tok/s | reported | 90 | - GB | 2,048 | vllm |
Context tested
The table above is the tested context for each exact rig and quantization cell. Empty fields remain empty rather than being filled with an estimate.