Particle.news
Get it on Google Play
Download on the App Store
7 ARTICLES
·
2mo ago

Zhipu Open-Sources GLM-5.3-Flash and Confirms Ox Alpha Was Its Test Model

The company published model weights with vendor benchmarks and low introductory pricing, a move that could cut inference costs and speed domestic deployment of frontier multimodal AI.

Image
Image
Image
Image

Overview

  • Zhipu confirmed late Wednesday that the anonymous Ox Alpha model used in public tests is GLM-5.3 and released the model as GLM-5.3-Flash with published weights and APIs.
  • The open release describes GLM-5.3-Flash as a native multimodal model with 320 billion parameters and 18 billion activation parameters designed to reduce runtime cost.
  • Zhipu says the model uses a mix of sparse and linear attention plus a manifold-constrained hyper-connection (mHC) design to preserve long-context ability while cutting layers and activations.
  • The company published vendor benchmarks that put GLM-5.3-Flash on par with Anthropic’s Claude Opus 4.8 on selected tests and announced steep introductory pricing that aims to undercut comparable offerings.
  • Moore Threads announced Day-0 support on Thursday, saying its MTT S5000 card and MUSA stack already run the model efficiently, which may speed real-world access on domestic accelerators and boost developer uptake.