Moonshot AI Challenges Proprietary Giants with Kimi K3, a 2.8T Open-Weight Model

Moonshot AI has launched Kimi K3, an open-weight AI model that challenges proprietary giants in the industry. The 2.8-trillion-parameter model is designed for coding and reasoning tasks, with a 1-million-token context window, the largest among open models.

Kimi K3 incorporates several architectural updates to tackle scaling challenges, including Kimi Delta Attention (KDA), Attention Residuals, and the Stable LatentMoE Framework. Benchmark results show Kimi K3 outperforming proprietary competitors in coding benchmarks like SWE Marathon and BrowseComp.

The model's pricing structure is aggressive, with API pricing at $0.30 per 1 million tokens for cache-hit inputs and $3.00 per 1 million tokens for cache misses. Output tokens are billed at $15.00 per 1 million tokens, incentivizing developers to optimize for cache efficiency.

Moonshot AI's release of Kimi K3 highlights the growing shift towards open-source alternatives in the AI landscape. The model's success could boost adoption of open-weight models across industries and set a precedent for future large-scale AI development.