Moonshot AI launched its latest model, Kimi 2 Thinking

Moonshot AI launched its latest model, Kimi 2 Thinking, positioning it as a direct competitor against market leaders like GPT-5 and Claude Sonnet 4.5.

The model is built on a one-trillion-parameter architecture with 32 billion active parameters during inference and context windows that exceed 250K tokens. It's open-sourced under a commercially friendly license, allowing researchers and companies to train and deploy it without significant restrictions.

Recent benchmarks show Kimi 2 Thinking competes with GPT-5 in reasoning, coding, and complex task planning, demonstrating how much China's AI ecosystem has advanced.

The development cycle is now quick enough to deliver competitive releases with global reach, supported by domestic compute and a strategic push for affordability.

Open weights, lower costs, and state-of-the-art performance mark a new phase of AI competition where access and infrastructure are as important as capability.