Moonshot released Kimi K3, a 2.8T-parameter MoE model with 104B active parameters, alongside its technical report and supporting infrastructure tools. Simultaneously, NVIDIA formally launched the Open Secure AI Alliance to promote shared tooling and research for AI security.
- Kimi K3 features native visual understanding, 1M-token context, and reports a ~2.5x scaling-efficiency improvement over K2.
- The release includes open-source components FlashKDA, MoonEP, and AgentENV, with distribution via vLLM, Baseten, and others.
- NVIDIA's alliance aims to ensure defenders have access to both open and closed frontier models for security purposes.
- Anthropic clarified its stance on open-weights, supporting chip controls and safety testing but not signing NVIDIA's letter.
- Kimi K3 ranks #1 among open-weight models in Agent Arena and #1 overall in Frontend Code Arena.
The release of Kimi K3 highlights the growing supply-chain nature of frontier model launches, while the alliance underscores increasing governance pressure on AI development.