Aikido fits security AI in 328 GB

Aikido released Altar, an open-weight security model that compresses GLM-5.3 from 1.51 TB to 328 GB and can run on a four-H200 node. Aikido’s release reports 60.4% average recall on a 32-vulnerability CVE benchmark, finding 23 of 32 vulnerabilities at least once across three runs; the full-precision parent averaged 65.6% and covered 25. The weights are available through Aikido’s organization, making air-gapped deployment possible without sending source code to a third-party inference service. Security teams should treat this as a workload-specific compression trade, not a drop-in frontier replacement, and compare it with the agent-evaluation boundary analysis before granting autonomous write access.