According to DeepLearning.AI, Moonshot AI has launched the Kimi K2 Thinking and Kimi K2 Thinking Turbo models, which leverage cycles of reasoning and tool use, making hundreds of calls to surpass other open-weights large language models (LLMs) on complex, multi-step tasks. These models are built as trillion-parameter mixture-of-experts architectures and fine-tuned at INT4 precision, resulting in strong agentic performance while maintaining compatibility with lower-cost hardware. This approach offers significant business opportunities for companies seeking scalable and cost-efficient AI solutions for advanced task automation and intelligent agent deployment (source: DeepLearning.AI, Nov 22, 2025).