LLM Inference Benchmark Explorer
Qwen3-4B-Instruct-2507 on 1× DGX Spark — NVFP4, vLLM, TP1 inference benchmark
Open this configuration in the LLM Inference Benchmark Explorer
| Model | Qwen3-4B-Instruct-2507 |
|---|---|
| Parameters | 4B |
| Intelligence Index | 8.8 |
| Agentic Index | — |
| Device | 1× DGX Spark |
| Quantization | NVFP4 |
| Max C | 32+ |
| Chat Capacity | 78 |
| Agentic Capacity | 19 |
| TP | — |
| DP | — |
| PP | — |
| Engine | vLLM |
| Speculative Decoding | — |
| C | TTFT (ms) | TPS (tok/s) | Status |
|---|---|---|---|
| 1 | 42 | 49.63 | PASS |
| 2 | 44 | 50.42 | PASS |
| 4 | 60 | 48.71 | PASS |
| 8 | 77 | 46.66 | PASS |
| 16 | 91 | 44.07 | PASS |
| 32 | 114 | 39.30 | PASS |
In OpenZeka's measurement, Qwen3-4B-Instruct-2507 (4B parameters), served in NVFP4 format with vLLM on 1× DGX Spark, reached a generation speed of 49.6 tok/s per request and a time to first token (TTFT) of 42 ms with a single request.
Considering the speed targets and the available KV cache capacity, the estimated capacity is 78 users for chat use and 19 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.
Intelligence Index and Agentic Index values are published by Artificial Analysis and are reproduced here with attribution.
In OpenZeka's measurement, Qwen3-4B-Instruct-2507 (4B parameters), served in NVFP4 format with vLLM on 1× DGX Spark, reached a generation speed of 49.6 tok/s per request and a time to first token (TTFT) of 42 ms with a single request.
Considering the speed targets and the available KV cache capacity, the estimated capacity is 78 users for chat use and 19 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.