Moonshot AI’s Kimi K3 is drawing attention as one of the largest open model releases to date. Latent Space describes it as a 2.8-trillion-parameter mixture-of-experts model with a 1M-token context window.
The model’s scale and open availability matter because frontier-style capabilities are increasingly appearing outside fully closed labs. If performance claims hold up in independent testing, Kimi K3 could influence how developers choose between open, hosted, and proprietary models.
The release also continues a broader shift in the market: open models are no longer just low-cost alternatives, but serious contenders for long-context and agentic workloads.