Benchmarks MiniMax M2.5

MiniMax M2.5

MiniMax • 456B MoE • Q5_K_M
28.4
tokens/sec

Best all-rounder for agentic tasks. Fast despite size. Our default for Polly and Scout.

agentsgeneralcoding

Specifications

Parameters
456B MoE
Active Parameters
45B
Quantization
Q5_K_M
VRAM Used
171GB
Context Length
32,768 tokens

Performance

Tokens/Second
28.4
Time to First Token
1.2s

Benchmark Scores

coding 8.5/10
reasoning 8.8/10
creative 8.2/10
agents 9/10
context 8.7/10

Quick Setup

Ollama
ollama run minimax-m2.5
LM Studio
Search 'MiniMax M2.5' in Discover
llama.cpp
llama-server -m MiniMax-M2.5-Q5_K_M.gguf -ngl 99