MiniMax Models
Chinese AI pioneer • MoE efficiency leader
5 Models
Latest: MiniMax-M2.5
MiniMax-M2.5
Released February 2026
MiniMax's flagship model matching frontier performance at 1/20th the cost with 80.2% SWE-bench Verified
Parameters
230B (10B active)
Context
200,000 tokens
Key Features
Frontier Performance Ultra Low Cost Open Weights (MIT)
+1 more
View Details →
MiniMax-M2.5-Lightning
Released February 2026
Ultra-fast variant of M2.5 generating 100 tokens per second at $1/hour continuous operation
Parameters
230B (10B active)
Context
200,000 tokens
Key Features
100 tok/s Speed $1/hour Operation Lightning Fast
+1 more
View Details →
MiniMax-M2.1
Released December 2025
Lightweight coding-focused model with strong multi-language programming and agentic capabilities
Parameters
230B (10B active)
Context
196,000 tokens
Key Features
Multi-language Coding Agentic Workflows Open Weights
+1 more
View Details →
MiniMax-M2
Released October 2025
MiniMax's general-purpose foundation model with strong tool use and deep search capabilities
Parameters
Not disclosed
Context
Not disclosed
Key Features
Open Source Tool Use Deep Search
+1 more
View Details →
MiniMax-M1
Released June 2025
World's first open-source large-scale hybrid-attention reasoning model with 1M token context
Parameters
456B (45.9B active)
Context
1,000,000 tokens
Key Features
1M Context Lightning Attention Open Source
+1 more
View Details →