LLM Inference Benchmark Explorer
GLM-5.3-Flash on 2× DGX Spark — NVFP4, vLLM, TP2 inference benchmark
Open this configuration in the LLM Inference Benchmark Explorer
| Model | GLM-5.3-Flash |
|---|---|
| Parameters | 321B |
| Intelligence Index | 41.8 |
| Agentic Index | 50.9 |
| Device | 2× DGX Spark |
| Quantization | NVFP4 |
| Max C | 1 |
| Chat Capacity | 2 |
| Agentic Capacity | 1 |
| TP | 2 |
| DP | — |
| PP | — |
| Engine | vLLM |
| Speculative Decoding | — |
| C | TTFT (ms) | TPS (tok/s) | Status |
|---|---|---|---|
| 1 | 395 | 24.34 | PASS |
| 2 | 575 | 17.66 | FAIL |
| 4 | 765 | 13.58 | FAIL |
| 8 | 4300 | 8.94 | FAIL |
| 16 | 18900 | 4.51 | FAIL |
In OpenZeka's measurement, GLM-5.3-Flash (321B parameters), served in NVFP4 format with vLLM on 2× DGX Spark (TP=2), reached a generation speed of 24.3 tok/s per request and a time to first token (TTFT) of 395 ms with a single request.
Considering the speed targets and the available KV cache capacity, the estimated capacity is 2 users for chat use and 1 for agentic use. The realistic capacity will likely fall between these two values. In scenarios dominated by coding, tool use, long workflows and multi-agent use, capacity approaches the agentic estimate; where shorter interactions, standard conversations and lighter tasks dominate, it approaches the chat estimate.
Notes: Model: LibertAIDAI/GLM-5.3-Flash-NVFP4. Docker: radixark/vllm-glm53-flash:sm121-v8
Intelligence Index and Agentic Index values are published by Artificial Analysis and are reproduced here with attribution.
In OpenZeka's measurement, GLM-5.3-Flash (321B parameters), served in NVFP4 format with vLLM on 2× DGX Spark (TP=2), reached a generation speed of 24.3 tok/s per request and a time to first token (TTFT) of 395 ms with a single request.
Considering the speed targets and the available KV cache capacity, the estimated capacity is 2 users for chat use and 1 for agentic use. The realistic capacity will likely fall between these two values. In scenarios dominated by coding, tool use, long workflows and multi-agent use, capacity approaches the agentic estimate; where shorter interactions, standard conversations and lighter tasks dominate, it approaches the chat estimate.