MiniMax
MiniMax-M2.5
Performancebenchmark · gemessen am 20.07.2026 11:25
Benchmark-ID
run-20260722-165904-a1f0cbPerformancetest Small 1.0MoE230BRuntime: godclawQuantisierung: BF16
Generation62,70tok/s
Prefill368,13tok/s
Time to First Token144,00ms
Gesamtdauer16,48s
Concurrency1parallel
Einordnung im Feld
28von 135 Systemen
Performancebenchmark · Leitmetrik: Generation-Speed (tok/s)
Dieser Lauf ist besser als 79 % aller vergleichbaren Systeme.
Generation
62,7 tok/s
+48 % vs Ø 42,5
Prefill
368,1 tok/s
-87 % vs Ø 2.885,5
Time to First Token
144 ms
-91 % vs Ø 1.685
Verteilung im Feld2 – 148 tok/s
Wie schlägt sich dieser Benchmark?
gpt-oss-20b3× AMD Radeon AI PRO R9700 · run-20260722-165909-85f64f
147,8 tok/s
gpt-oss-20b3× AMD Radeon AI PRO R9700 · run-20260722-165909-637af7
143,9 tok/s
gpt-oss-20b3× AMD Radeon AI PRO R9700 · run-20260722-165909-65d35f
137,7 tok/s
Nemotron-Cascade-2-30B-A3B3× AMD Radeon AI PRO R9700 · run-20260722-181156-a273f2
107,9 tok/s
Nemotron-3-Nano-Omni-30B-A3B-Reasoning3× AMD Radeon AI PRO R9700 · run-20260722-182228-95c528
102,9 tok/s
gpt-oss-120b3× AMD Radeon AI PRO R9700 · run-20260722-165909-4e3ede
99,8 tok/s
Qwen3-30B-A3B-Instruct-25073× AMD Radeon AI PRO R9700 · run-20260722-165907-cc8f3f
97,6 tok/s
NVIDIA-Nemotron-3-Nano-4B3× AMD Radeon AI PRO R9700 · run-20260722-190202-45dc8e
93,4 tok/s
Qwen3-30B-A3B-Thinking-25073× AMD Radeon AI PRO R9700 · run-20260722-165908-5712c8
91,4 tok/s
Qwen3-30B-A3B-Instruct-25073× AMD Radeon AI PRO R9700 · run-20260722-165907-b20fe5
91,0 tok/s
Qwen3-VL-30B-A3B-Instruct3× AMD Radeon AI PRO R9700 · run-20260722-165908-4d91c3
89,8 tok/s
Qwen3-30B-A3B-Thinking-25073× AMD Radeon AI PRO R9700 · run-20260722-165908-40e052
88,6 tok/s
Qwen3-Coder-30B-A3B-Instruct3× AMD Radeon AI PRO R9700 · run-20260722-165907-0a4b43
88,2 tok/s
Qwen3-30B-A3B-Thinking-25073× AMD Radeon AI PRO R9700 · run-20260722-165907-83951d
87,8 tok/s
MiniMax-M2.5 dieser Lauf3× NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition · run-20260722-165904-a1f0cb
62,7 tok/s
Hardware
GPU: 3x NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition · 96 GB VRAM
CPU: 34x AMD Ryzen Threadripper PRO 9965WX 24-Cores
RAM: 125 GB
Setup
Runtime: godclaw
Quantisierung: BF16
Modell: MiniMax-M2.5
Konfiguration
# LLM-Benchmark Konfiguration
# Modell : MiniMax-M2.5
# Engine : vLLM
# Run-ID : run-20260722-165904-a1f0cb
# GPU : 3x NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition
# CPU : 34x AMD Ryzen Threadripper PRO 9965WX 24-Cores
# RAM : 125 GB
bench@llm-benchmark:~$ /usr/bin/python3 /usr/local/bin/vllm serve \
--model mratsim/MiniMax-M2.5-BF16-INT4-AWQ \
--served-model-name MiniMax-M2.5 \
--tensor-parallel-size 1 \
--pipeline-parallel-size 3 \
--max-model-len 164000 \
--gpu-memory-utilization 0.90 \
--load-format safetensors \
--trust-remote-code \
--enforce-eager \
--disable-custom-all-reduce \
--enable-auto-tool-choice \
--tool-call-parser minimax_m2 \
--reasoning-parser minimax_m2_append_think
| Engine | vllm |
| Modellalias | MiniMax-M2.5 |
| Kontextlaenge | 164000 |
| Modellpfad | mratsim/MiniMax-M2.5-BF16-INT4-AWQ |
| Alias | MiniMax-M2.5 |
| Tensor-Parallel | 1 |
| Pipeline-Parallel | 3 |
| Kontext | 164000 |
| GPU-Speicher | 0.90 |
| Load-Format | safetensors |
| Trust-Remote-Code | aktiv |
| Enforce-Eager | aktiv |
| Custom-AllReduce aus | aktiv |
| Auto-Tool-Choice | aktiv |
| Tool-Parser | minimax_m2 |
| Reasoning-Parser | minimax_m2_append_think |