Kimi K3 debate centers on a claimed 2.8-trillion-parameter MoE model
pstAsiatech · x · 2026-07-24
The quoted thread pushes back on the idea that Kimi K3 is merely a “cloned” model.
It argues that pretraining a 2.8-trillion-parameter foundation model with sparse MoE routing is an engineering feat that cannot be faked, and says the real question is what the lab has actually achieved at scale.
More from Models
- AutoCAD-Bench says GPT-5.6 Sol leads computer-use CAD tasks with 46% — gabrielchua · 2026-07-24
- Moonshot’s Kimi K3 shows how open models can turn outside compute into an advantage — scientificamerican · 2026-07-24
- Claude keeps saying it’s tired, and users are calling the act out — econoar · 2026-07-24
- Kimi k3 looks strong, but benchmark scores still don’t prove real-world quality — FuSheng_0306 · 2026-07-24
- Apertus 1.5 goes public with chat access and image capabilities — valentina__py · 2026-07-24
- Claude Opus 4.8 is too verbose, while GPT-5.6 feels sharper — santoshpanda · 2026-07-24