GitHub radar
Kimi K3: 1M Context Multimodal via Open API
Moonshot AI's Kimi K3 brings 2.8T-parameter multimodal intelligence and a 1-million-token context window to an OpenAI-compatible API.
Kimi K3 is Moonshot AI's flagship multimodal model, released in June 2026. It uses a Mixture-of-Experts architecture with 2.8 trillion total parameters, activating 104 billion per token (896 experts, 16 active per token). A built-in MoonViT-V2 vision encoder (401M parameters) enables native image understanding. Context window: 1,048,576 tokens — roughly 750K words in a single request. The new Kimi Delta Attention architecture delivers a claimed 2.5× improvement in scaling efficiency over K2. Strong benchmark results: Terminal-Bench 2.1 scores 88.3 (coding) and Harvey Lab-AA 94.6 (legal reasoning). Supports self-hosted deployment via vLLM and SGLang, plus a cloud API at platform.kimi.ai with OpenAI- and Anthropic-compatible endpoints.
Why a vibe-coder should care
A million-token context is rare: it lets you analyze entire codebases or long documents in one request without splitting. Kimi K3 is accessible via an open API with OpenAI-format compatibility — connect to an agent or IDE in minutes. For deep text and image processing tasks, it is one of the strongest options available through an open API.
How to install
Copy this and send it to your agent — Claude Code, Codex, any of them:
Set up Kimi K3 API access from Moonshot AI: https://platform.kimi.ai — get me an API key, show a sample OpenAI-format request, and explain how to pass an image in the request.
Server-grade model — 2.8 trillion parameters, you can't run this at home. Available via API at platform.kimi.ai with no infrastructure setup.
Open on Hugging Face▌ More finds