MiniMax M2.5 Review: Open-Weight Agentic LLM — 229B MoE, 80.2% SWE-Bench, BFCL Leader, $1.15/M Output

MiniMax M2.5 (February 12, 2026) is the Chinese AI company's open-weight flagship: 229 billion total parameters, 10 billion active (Sparse MoE), 200K context, and a Forge RL framework trained across 200,000+ real-world environments. We review the architecture, benchmark claims, pricing, controversies, and whether the price-performance ratio justifies the hype.