Aikido Security Releases Altar-1: An Open-Weight Security Model Pruned From GLM-5.3 to 328 GB
328 GB. Aikido Security compressed a frontier model to fit air-gapped networks—and proved you don't need cloud for security AI.
Why it matters
Aikido released Altar-1, an open-weight pruned model derived from GLM-5.3, deployable on-prem for autonomous pentesting. The move signals a shift toward locally-controllable security models, but the novelty lies in the engineering (pruning, vLLM compatibility) rather than new capability—and the base model (GLM-5.3) is neither frontier nor newly released.
The key facts
14 to knowAltar-1 is an open-weight pruned derivative of Z.AI's GLM-5.3
Model size: 328 GB
Deployable on-prem and air-gapped networks via vLLM
Weights published on Hugging Face
Powers Aikido Machine, an autonomous pentesting appliance
No capability benchmarks or evals disclosed in summary
No comparison to uncompressed GLM-5.3 performance or other security models
Model: Altar-1, pruned from Z.AI's GLM-5.3
Size: 328 GB (compressed/deployable open-weight)
Distribution: weights public on Hugging Face
Runtime: vLLM-compatible
Use case: Aikido Machine autonomous pentesting appliance
Deployment scope: on-prem and air-gapped networks
Status: weights released; no capability benchmarks, attack accuracy metrics, or comparative eval vs proprietary security models disclosed
The story so far
Earlier coverage of this storyline
- Xiaomi introduces Mimo-V2.6 series open-source AI model familySiliconAngle
- This story
Go to the source
MarkTechPostmarktechpost.com
Publisher excerpt: Aikido Security has released Altar-1, its first open-weight security model. It is a compressed version of Z.AI’s GLM-5.3, built to run inside infrastructure the customer controls. Altar-1 powers Aikido Machine, the company’s autonomous pentesting appliance for on-prem and air-gapped networks. Is it…