Moonshot AI

Generative AI · Cloud Infrastructure · Open-Weight Models

Launch

Moonshot AI's 2.8T-parameter Kimi K3 goes live on Amazon Bedrock

It runs inside the same security boundary as proprietary systems, erasing the tradeoff between open-weight flexibility and enterprise governance.

• Kimi K3, Moonshot AI's 2.8-trillion-parameter open-weight model, became generally available on Amazon Bedrock on September 18, 2026, across all AWS Regions where Bedrock operates.
• Moonshot AI built Kimi K3 on a mixture-of-experts design it calls Stable LatentMoE, activating 16 of 896 experts, plus new attention mechanisms it terms Kimi Delta Attention and Attention Residuals.
• The model pairs native vision with a 1-million-token context window and is the first open-weight model on Bedrock to support explicit prompt caching, cutting latency and input costs on reused context.
• Bedrock offers Kimi K3 through a global cross-region inference profile that is roughly 10% cheaper than a US-only, data-residency-compliant regional profile.
• Moonshot AI says Kimi K3 delivers roughly 2.5x better scaling efficiency than its predecessor Kimi K2, but acknowledges the model still trails proprietary rivals Claude Fable 5 and GPT 5.6 Sol on overall performance.
• Bedrock has added dozens of open-weight models since 2025 from DeepSeek, Google, MiniMax, Mistral AI, NVIDIA, OpenAI, and Qwen, backed by 2026 platform upgrades like tool calling, reasoning, and streaming.
• By running open-weight models inside the same access, encryption, and auditing controls used for proprietary systems, cloud platforms are dissolving the old choice enterprises faced between open-model flexibility and governance rigor.