MiniMax-M2.5
Built for Real-World Productivity
#Open Source
#Software Engineering
#Artificial Intelligence
#GitHub
MiniMax-M2.5 – Open-source model optimized for real-world productivity
Summary: MiniMax-M2.5 is an open-source frontier model delivering state-of-the-art performance in coding, search, agentic tool-calling, and office tasks. It executes complex tasks 37% faster and supports scalable long-horizon agents at $1 per hour with 100 transactions per second.
What it does
It provides high accuracy in coding (SWE-Bench 80.2%), search (BrowseComp 76.3%), and agentic tool-calling (BFCL 76.8%), optimized for efficient execution and scalability.
Who it's for
Developers and organizations needing an open-source model for coding, search, and automated workflows with scalable performance.
Why it matters
It enables economically viable, scalable deployment of long-horizon agents with competitive performance against proprietary models.