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 →
| Model | Engine | Hardware | tok/s | Metric | Source level | Date | Source |
|---|---|---|---|---|---|---|---|
| MiniMax-M3.1-Flash | SGLang | 4× DGX Spark (TP4, switchless RoCE ring) | 59 | SGLang 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 profile | Measured | 2026-10-25 | link |
| MiniMax-M3.1-Flash | NVIDIA Flare | 4× DGX Spark | 45 | C1 45; aggregate (separate) C2 61 / C4 75. The author says Flare is not fully optimized yet | Measured | 2026-10-27 | link |
| MiniMax-M3.1-Flash | Project Maya | RTX 5090 32GB + 92GiB RAM | 24 | Maya 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 128K | Measured | 2026-11-03 | link |
| MiniMax-M3.1-Flash | vLLM | 2× DGX Spark | 21.6 | Code 21.6 (2.15bpw gives 29.1); IFEval 74.5% vs 60.5%; decode stays flat to 146K | Measured | 2026-10-28 | link |
| MiniMax-M3.1-Flash | Strata (syoyo fork) | 2× Tesla V100 32GB + 181GB RAM | 40.2 | syoyo 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 512K | Claimed | 2026-11-04 | link |
| MiniMax-M3.1-Flash | llama.cpp | Mac Studio M5 Ultra | 40 | llama.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 end | Claimed | 2026-11-04 | link |
| MiniMax-M3.1-Flash (abliterated, gate/up Q3) | ds4 (DwarfStar) | Mac Studio M3 Ultra 256GB | 31 | Greedy 31, long text 19.5. tok_s uses greedy 31 | Claimed | 2026-11-05 | link |
| MiniMax-M3.1-Flash | Project Maya | 4× RTX 3090 24GB + EPYC (2 of 8 memory channels filled) | 23 | Decode 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/s | Claimed | 2026-10-28 | link |
| MiniMax-M3.1-Flash | Unspecified | 2× DGX Spark | 23 | Code 24.7 / prose 23.0; TTFT at 8K 17.7s. tok_s uses prose 23.0 | Claimed | 2026-11-05 | link |
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.