# @themacrosift on Instagram

- **Type:** Image
- **Original URL:** https://www.instagram.com/p/DbTrKkFB9zm
- **Gondola URL:** https://gondola.cc/posts/68552306-themacrosift-instagram
- **Thumbnail:** https://img.gondola.cc/tr:w-,h-,fo-auto/postThumbnails/cb1fb19d5d.jpg
- **Posted:** 2026-07-27T19:05:56.000+00:00
- **Account Owner:** The Macro Sift | Exponential Tech (@themacrosift) — https://gondola.cc/themacrosift

## Caption

The largest open weight AI model ever released is here.

Moonshot AI has published the full weights for Kimi K3, a 2.8 trillion parameter mixture of experts model that immediately becomes the largest openly available model released to the public.

What makes it interesting is not just its size.

K3 uses an MoE architecture with 896 experts, but activates only 16 experts per token, dramatically reducing inference costs while maintaining trillion scale capability.

Other highlights include:

• 1 million token context window
• Native multimodal capabilities
• API pricing around $3 per million input tokens and $15 per million output tokens
• Strong early benchmark performance, including a top finish in Frontend Code Arena

The benchmarks are impressive, but the real validation starts now that researchers and developers can inspect, test, and build on the released weights.

One important distinction deserves attention.

Open weights are not the same as universal accessibility. Running Kimi K3 efficiently still requires approximately 64 AI accelerators, making self hosting impractical for most organizations and individuals.

The competition between open and closed frontier models is accelerating, and each new release is making advanced AI capabilities more widely available, even if the underlying infrastructure remains expensive.

#Kimi #OpenSource #AI #MacroSift

## Stats

- **Views:** 0
- **Likes:** 2
- **Shares:** 0
- **Comments:** 0

## Tags

kimi, macrosift, ai, opensource

---
Copyright (c) Gondola