Kimi-K3 release narrows the gap between open-source and closed-source AI models
The release of Kimi-K3, a 2.8-trillion-parameter open-source model, has brought the open-source AI frontier to within 1.5 months of leading closed-source models. This development demonstrates that scaling laws continue to hold for larger models, significantly narrowing the gap between open-source alternatives and proprietary giants like OpenAI and Anthropic. Built on Kimi Delta Attention and Attention Residuals, Kimi-K3 features native vision capabilities, a 1-million-token context window, and is optimized for long-horizon coding and knowledge work.
## BACKGROUND
Large language models rely on scaling parameters to improve performance, but training and running these multi-trillion parameter models requires massive computational resources. Independent platforms like Artificial Analysis benchmark these models to track performance gaps between proprietary and open-source AI.