MiniMax, an international leader in artificial intelligence innovation, announced the launch of MiniMax M2.5, its latest frontier AI model engineered for real-world productivity across coding, agentic workflows, search, and professional office tasks. The new model delivers state-of-the-art performance while dramatically lowering operating costs, enabling broad access to intelligent automation.
Extensively trained using reinforcement learning across hundreds of thousands of complex environments, MiniMax M2.5 sets new performance benchmarks with significant improvements in speed, reasoning, and cost efficiency positioning itself as a competitive alternative at the forefront of AI model development.
Industry-Leading Multidimensional Performance
MiniMax M2.5 achieves strong performance across diverse benchmarks, including:
-
Coding: The model scored 80.2% on SWE-Bench Verified, showcasing robust engineering and software development capabilities.
-
Multitask Reasoning and Search: It delivered 51.3% on Multi-SWE-Bench and 76.3% on BrowseComp with contextual management.
-
Speed: MiniMax M2.5 performs complex evaluations 37% faster than its predecessor, M2.1, and matches the inference speed of comparable leading models.
With enhanced task decomposition and optimized agentic reasoning, M2.5 plans like a software architect from system design to feature expansion and comprehensive testing across full-stack environments that include web, mobile, and server logic.
Also Read: ABBYY Launches Vantage 3.0 with GenAI and Compliance Features
“Built for Real-World Productivity”
MiniMax describes the philosophy behind the new model succinctly:
“Today we’re introducing our latest model, MiniMax-M2.5.”
Designed to excel in complex, commercially valuable tasks, the model supports a comprehensive range of real-world scenarios including search, autonomous tool use, detailed information retrieval, and professional workflows including Word document formatting, PowerPoint creation, and advanced Excel financial modeling.
Unbeatable Cost Efficiency for AI-Driven Workloads
MiniMax M2.5 significantly reduces the cost of deploying powerful AI:
-
Running the model at 100 tokens per second costs just $1 per hour, while operations at 50 tokens per second drop to $0.30 per hour.
-
These cost efficiencies are designed to make advanced AI capabilities scalable and economically practical for developers, enterprises, and production applications.
MiniMax’s strategic focus is to provide intelligence that is “too cheap to meter”, supporting innovation without traditional cost barriers.
Source: MiniMax
