⚡ Local Inference Recipes

English | 简体中文 | 日本語

MiniMax: local inference speed (tok/s), sourced

As of 2026-11-14, this dataset has 9 recipes for MiniMax. 4 were measured by the poster. Fastest measured single-device, single-stream text result: 24 tok/s with MiniMax-M3.1-Flash on Project Maya (RTX 5090 32GB + 92GiB RAM; Measured, 2026-11-03; source). 9 entries have a date, from 2026-10-25 to 2026-11-05.

Filter these in the explorer →

ModelEngineHardwaretok/sMetricSource levelDateSource
MiniMax-M3.1-FlashSGLang4× DGX Spark (TP4, switchless RoCE ring)59SGLang TP4 / nnodes=4, NCCL 2.30.7 Ring, fp8 KV: single-stream decode 59.0 (TTFT 0.183 s); prefill 32K 1,121 / 266K 1,248; aggregate (separate) 6 concurrent 122.9. Recipe from the knapcio SGLang profileMeasured2026-10-25link
MiniMax-M3.1-FlashNVIDIA Flare4× DGX Spark45C1 45; aggregate (separate) C2 61 / C4 75. The author says Flare is not fully optimized yetMeasured2026-10-27link
MiniMax-M3.1-FlashProject MayaRTX 5090 32GB + 92GiB RAM24Maya 1.1.50 default install, median of 2 runs of 512 tokens, thinking off: decode 24.0 (code 24.4 / prose 23.6); 18.7 with a 249K prompt. TTFT 36.6 s at 16K, 465.8 s at 128KMeasured2026-11-03link
MiniMax-M3.1-FlashvLLM2× DGX Spark21.6Code 21.6 (2.15bpw gives 29.1); IFEval 74.5% vs 60.5%; decode stays flat to 146KMeasured2026-10-28link
MiniMax-M3.1-FlashStrata (syoyo fork)2× Tesla V100 32GB + 181GB RAM40.2syoyo Strata fork: 2× V100 decode 40.2 (exact) / 65.2 (lossy --speed max); 1× V100 23–28 (lossy 56); prefill 634–640 at 16K. Decode falls at long context: 12.8 at 128K, 6.0 at 512KClaimed2026-11-04link
MiniMax-M3.1-Flashllama.cppMac Studio M5 Ultra40llama.cpp tuned by the author (Metal kernel and server changes, KLD unchanged): single session with DSpark 40–60; prefill at 128K 817 → 945; 8-session aggregate (separate) about 59 → about 110. tok_s uses the low endClaimed2026-11-04link
MiniMax-M3.1-Flash (abliterated, gate/up Q3)ds4 (DwarfStar)Mac Studio M3 Ultra 256GB31Greedy 31, long text 19.5. tok_s uses greedy 31Claimed2026-11-05link
MiniMax-M3.1-FlashProject Maya4× RTX 3090 24GB + EPYC (2 of 8 memory channels filled)23Decode 23–31 (best llama.cpp on the same machine 6.7); prefill 16K 1,120 / 28K 1,525 (llama.cpp 219); cold experts through the CPU at about 19 GB/sClaimed2026-10-28link
MiniMax-M3.1-FlashUnspecified2× DGX Spark23Code 24.7 / prose 23.0; TTFT at 8K 17.7s. tok_s uses prose 23.0Claimed2026-11-05link

Numbers come from each source. This project did not measure them again. Models and settings differ between rows. Source levels: Official = project or vendor; Measured = the poster ran it (an engine or recipe author must also publish raw data and reproduction steps); Relayed = second-hand; Claimed = prediction or no reproduction details.