MiniMax-M3 데뷔, 주요 벤치마크 성능에서 GPT-5.5와 Gemini 3.1 Pro를 능가하며 비용은 단 5-10% 수준

MiniMax-M3 challenges US AI dominance with an open-weights model that outperforms GPT-5.5 and Gemini 3.1 Pro on coding benchmarks while costing 5-10% of rivals—a breakthrough enabled by Sparse Attention architecture that cuts computational overhead by 20x. For enterprises, local deployment eliminates cloud vendor lock-in and data exposure risks, particularly critical in regulated industries. This demonstrates that non-US players can achieve performance parity through architectural innovation while offering dramatically superior economics.