DigestAI news desk

Cut through the AI noise.

Research4 min read

Aikido Security releases Altar-1: a 328 GB security model from GLM-5.3

Aikido Security launched Altar-1, its first open-weight security model derived from Z.AI’s GLM-5.3. The model compresses GLM-5.3’s 753B-parameter mixture-of-experts architecture to 328 GB by pruning 34.4% of experts and quantizing weights to 4 bits. It retains 168 of 256 routed experts per layer, with 8 active per token, and runs on a single node of 4x NVIDIA H200 GPUs using vLLM.

1 source

Key points

  • Aikido Security pruned GLM-5.3 to 328 GB, a 78.2% reduction from its parent model’s size
  • Altar-1 runs on 4x NVIDIA H200 GPUs with vLLM, supporting air-gapped networks
  • Recall drops from 65.6% to 60.4% while maintaining 92% of GLM-5.3’s CVE coverage

Altar-1 powers Aikido Machine, an autonomous pentesting tool for on-prem and air-gapped networks. Aikido claims it achieves 92% of GLM-5.3’s CVE coverage with a 1-point recall drop, based on an internal benchmark of 32 known vulnerabilities. The model also found a critical-severity vulnerability during a client’s production pentest. Deployment requires Hopper GPUs, and the GLM-5.3 license permits commercial use with a review clause for operators exceeding $10B in annual revenue.

Model page: Altar-1 →

Full story from MarkTechPost · by Asif RazzaqOpen source ↗

Aikido Security Releases Altar-1: An Open-Weight Security Model Pruned From GLM-5.3 to 328 GB

MarkTechPost · 25 September 2026

Loading the full article…

This text was published by MarkTechPost and written by Asif Razzaq. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗

Topics · follow one to build your own front page
Aikido SecurityZ.AIAltar-1GLM-5.3

The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.

Comments

via GitHub Discussions

More in Research

All →

Related stories