GitHub radar
Kimi K3: MoonshotAI's Open-Weight Frontier Model
MoonshotAI released the weights of Kimi K3, its 2.8-trillion-parameter multimodal agentic model with a 1-million-token context window. It uses a Mixture-of-Experts architecture and is compatible with OpenAI and Anthropic API interfaces.
Kimi K3 is the open-weight flagship model from MoonshotAI, a Chinese AI company. It uses a Mixture-of-Experts architecture with 2.8 trillion total parameters and 16 active experts per token out of 896. The model processes up to 1 million tokens in a single context and supports both text and image input via MoonViT-V2 vision encoding. Strong benchmark results include GPQA Diamond 93.5 and MCPMark-Verified 94.5.
Why a vibe-coder should care
You can connect Kimi K3 to your project as a drop-in replacement for closed API models — it is compatible with OpenAI and Anthropic API interfaces, so no code changes are needed. Hosting it yourself via vLLM or SGLang gives you a capable multimodal agentic model without paying per-token to a third party.
▌ More finds