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

Aikido Security has released Altar-1, its first open-weight security model, compressed from Z.AI's 753B-parameter GLM-5.3 MoE to 328 GB. AWQ INT4 quantization and REAP expert pruning (168 of 256 experts per layer) cut size by 78.2%; it runs on a single node of 4x Nvidia H200 GPUs with vLLM, including air-gapped setups.

• Compressed from 1,506.7 GB to 328 GB, a 78.2% reduction vs BF16
• REAP keeps 168 of 256 experts per layer, 8 active per token
• CVE recall drops from 65.6% to 60.4%, retaining 92% of parent coverage
• Requires 4x H200; a 4x H100 80 GB node (320 GB) is too small