LLM Inference Benchmark Explorer
MiniMax-M3.1-Flash on 4× DGX Spark — NVFP4, vLLM, TP4, speculative decoding k=8 inference benchmark
Open this configuration in the LLM Inference Benchmark Explorer
| Model | MiniMax-M3.1-Flash |
|---|---|
| Parameters | 428B |
| Intelligence Index | — |
| Agentic Index | — |
| Device | 4× DGX Spark |
| Quantization | NVFP4 |
| Max C | 2 |
| Chat Capacity | 8 |
| Agentic Capacity | 3 |
| TP | 4 |
| DP | — |
| PP | — |
| Engine | vLLM |
| Speculative Decoding | Yes |
| C | TTFT (ms) | TPS (tok/s) | Status |
|---|---|---|---|
| 1 | 489 | 34.96 | PASS |
| 2 | 674 | 20.51 | PASS |
| 4 | 815 | 14.01 | FAIL |
| 8 | 1553 | 9.76 | FAIL |
In OpenZeka's measurement, MiniMax-M3.1-Flash (428B parameters), served in NVFP4 format with vLLM and speculative decoding on 4× DGX Spark (TP=4), reached a generation speed of 35 tok/s per request and a time to first token (TTFT) of 489 ms with a single request.
Considering the speed targets and the available KV cache capacity, the estimated capacity is 8 users for chat use and 3 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: DSpark k=8, 1M context. MXFP8 dense layers, NVFP4 experts. Measured up to 8 concurrent requests. Model: MiniMaxAI/MiniMax-M3.1-Flash
Intelligence Index and Agentic Index values are published by Artificial Analysis and are reproduced here with attribution.
In OpenZeka's measurement, MiniMax-M3.1-Flash (428B parameters), served in NVFP4 format with vLLM and speculative decoding on 4× DGX Spark (TP=4), reached a generation speed of 35 tok/s per request and a time to first token (TTFT) of 489 ms with a single request.
Considering the speed targets and the available KV cache capacity, the estimated capacity is 8 users for chat use and 3 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.