MiniMax-M2.5

MiniMax🇨🇳 China
active

MiniMax's latest open-weight reasoning model with improved coding and agentic task performance.

Context window197K tokens
Input / 1M tokens$0.3
Output / 1M tokens$1.1

Version History

2.5minor

M2.5 is trained specifically for agent-native execution using reinforcement learning on agentic scaffolds, with emphasis on tool-calling, multi-step task decomposition, and long-horizon coding tasks.

minimax-m2-5-launchmajor

MiniMax-M2.5 launches.

Benchmark Scores

Full leaderboard →
1391.0 elo
Arena Elo
84.8%
GPQA
89.3%
Hallucination Rate
82.0 tokens_per_sec
Speed (tok/s)
80.2%
SWE-bench Verified

Coverage