Moonshot AI Unveils Groundbreaking 2.8T-Parameter Open-Weight Model

Moonshot AI has unveiled Kimi K3, a significant milestone in the development of open-weight AI models. With its 2.8-trillion-parameter architecture, Kimi K3 surpasses the capabilities of many proprietary models, including OpenAI's GPT 5.6 Sol and Anthropic's Claude Fable 5.

Kimi K3 is designed for coding, long-context reasoning, and multimodal tasks, with a standout feature being its 1-million-token context window. This allows the model to process extensive knowledge and reasoning capacities across various applications.

The model incorporates several architectural updates, including Kimi Delta Attention (KDA) and Stable LatentMoE Framework, which enable efficient processing of long-context inputs and optimize performance while keeping compute costs manageable.

According to Moonshot AI, Kimi K3 outperforms proprietary competitors in several areas, particularly on coding benchmarks. The model's pricing structure is also aggressive, with API pricing set at $0.30 per 1 million tokens for cache-hit inputs and $3.00 per 1 million tokens for cache misses.