LLM Inference Benchmark Explorer

MiniMax-M3 on 4× DGX Spark — NVFP4, vLLM, TP4, speculative decoding k=8 inference benchmark

Open this configuration in the LLM Inference Benchmark Explorer

ModelMiniMax-M3
Parameters427B
Intelligence Index29.2
Agentic Index29.5
Device4× DGX Spark
QuantizationNVFP4
Max C2
Chat Capacity8
Agentic Capacity3
TP4
DP—
PP—
EnginevLLM
Speculative DecodingYes
CTTFT (ms)TPS (tok/s)Status
145334.49PASS
265325.21PASS
487415.61FAIL
886508.17FAIL

In OpenZeka's measurement, MiniMax-M3 (427B parameters), served in NVFP4 format with vLLM and speculative decoding on 4× DGX Spark (TP=4), reached a generation speed of 34.5 tok/s per request and a time to first token (TTFT) of 453 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: FLASHINFER_CUTLASS, DSpark k=8

Intelligence Index and Agentic Index values are published by Artificial Analysis and are reproduced here with attribution.