M
MiniMax M2.7
FreemiumChatChatCodeResearch
MiniMax M2.7 is a next-generation self-evolving large language model released on March 18, 2026. It can autonomously perform 30–50% of reinforcement learning research workflows and shows significant gains over M2.5 in software engineering, professional office tasks, and agentic benchmarks. Available via MiniMax Agent platform and API, M2.7 uses agent harnesses and continuous reinforcement learning to improve its own capabilities over time.
