Тестирование Minimax M3: производительность, кодинг и границы возможностей модели

The presented materials describe the launch of MiniMax M3, a new open-weights AI model that demonstrates outstanding results in programming and multimodal tasks. Thanks to its innovative sparse attention architecture (MSA), the model supports context windows of up to 1 million tokens and is significantly faster and cheaper than many closed-loop alternatives. In practical tests, M3 successfully generates complex SVG images, interactive web applications, and even 3D simulations, often outperforming giants like GPT-5.5 and Gemini. The sources emphasize the system's ability to autonomously reason and perform multi-step tasks without human intervention. Furthermore, the model's high availability via API and extremely aggressive pricing policy make cutting-edge technology accessible to a wide range of developers are noted. Thus, the release of MiniMax M3 is seen as a serious challenge to proprietary AI solutions.