Moonshot AI launches 2.8T parameter Kimi K3
Moonshot AI has launched Kimi K3, a 2.8-trillion-parameter multimodal Mixture-of-Experts reasoning model with a 1-million-token context window. The model features proprietary hybrid attention mechanisms and plans for an open-weights release by late July 2026.
Kimi K3 is a massive challenge to proprietary closed-source reasoning giants, combining a colossal 2.8T parameter MoE architecture with an open-weights release strategy that could democratize state-of-the-art agentic reasoning.
- –Ultra-large-scale parameter count: At 2.8T parameters, Kimi K3 is one of the largest models to commit to an open-weights release.
- –Hybrid attention innovations: Utilizing Kimi Delta Attention and Attention Residuals, the model seeks to optimize the compute efficiency of its 1M-token context window.
- –Shifting developer loyalty: Early users report that Kimi K3's coding and agentic capabilities make it a viable alternative to other frontier reasoning models, reinforcing the momentum of open-source AI.
DISCOVERED
1h ago
2026-07-16
PUBLISHED
2h ago
2026-07-16
RELEVANCE
AUTHOR
Av1dlive