The MiniMax M2.7 model is now publicly available through the MiniMax Agent platform and the MiniMax API Platform, the company said on March 19, 2026. The release marks a shift toward AI systems that participate in their own development cycle through agent harnesses and reinforcement learning.
M2.7 uses reinforcement learning to optimize its own capabilities, including memory updates, skill development, and iterative self-improvement. The model supports complex workflows across software engineering, office productivity, and research environments.
MiniMax M2.7 Model Performance on Key Benchmarks
The MiniMax M2.7 model achieves a score of 56.22% on the SWE-Pro benchmark and 55.6% on VIBE-Pro. These scores place it near the top of publicly available models on both evaluations. The model also records an ELO score of 1495 on GDPval-AA, outperforming most open-source models in professional office tasks.
Furthermore, M2.7 demonstrates a 97% skill adherence rate across more than 40 complex skills. This figure reflects the model’s consistency when executing specialized tasks in professional settings.
Technical Capabilities and Multi-Agent Collaboration
M2.7 introduces multi-agent collaboration, enabling multiple AI instances to coordinate on a single task. The model also supports autonomous debugging, research agent harnesses, and an OpenRoom demo designed for entertainment and interactive experiences.
Compared to its predecessors, M2.7 shows improvements in multilingual programming, code security, end-to-end project delivery, and deep system-level understanding. Early user feedback indicates that the model produces outputs that integrate directly into professional workflows.
Internal Deployment and AI-Native Strategy
MiniMax has deployed M2.7 internally to automate and optimize its own research and development processes. The company said the deployment requires minimal human intervention, reflecting its broader strategy to operate as an artificial intelligence-native organization.
This internal use of M2.7 positions the model as both a product and an operational tool. Consequently, MiniMax accelerates its own development cycles using the same system it offers to external developers.
Availability and Access
Developers and organizations can access M2.7 through the MiniMax Agent interface or directly via the MiniMax API Platform. The model targets use cases in software development, enterprise productivity, and applied research.
MiniMax has not disclosed pricing details for API access in the current release. However, the company said the platform supports both individual developers and enterprise-scale deployments. The release of M2.7 follows MiniMax’s ongoing effort to expand its model lineup with systems capable of handling end-to-end professional tasks with reduced human oversight.
“M2.7 supports complex workflows in software engineering, office productivity, and research environments, with a 97% skill adherence rate across 40+ complex skills.”
MiniMax, Company Statement