Moonshot AI has released Kimi K2.7 Code, an open-weight 1-trillion parameter Mixture-of-Experts coding model featuring native vision support and a 256K context window.
Moonshot AI has launched Kimi K2.7 Code, a 1-trillion parameter coding-focused Mixture-of-Experts (MoE) model with 32 billion active parameters. The model introduces native vision support, operates with a 256K context window, and reduces thinking token usage by 30% compared to Kimi K2.6, making it highly efficient for long-context programming and reasoning tasks.
Moonshot AI is successfully pushing the boundaries of open-weight developer tools by combining massive context windows with highly optimized reasoning token usage.
* The 30% reduction in thinking tokens compared to Kimi K2.6 lowers latency and compute costs during complex coding tasks.
* Integrating native vision support equips the model to handle multimodal workflows like UI styling, flowchart analysis, and visual debugging.
* Utilizing a Mixture-of-Experts architecture with 32B active parameters provides a strong balance between high model capacity and practical run-time compute requirements.
DISCOVERED
5d ago
2026-06-12
PUBLISHED
5d ago
2026-06-12
RELEVANCE
AUTHOR
Bijan Bowen