Aikido Launches Altar Open-Weight Model for On-Prem Security Workflows
Summary
Aikido says its Altar model compresses GLM-5.3 to 328 GB for on-prem security workflows. On an internal 32-vulnerability benchmark, it averaged 60.4% recall; the company notes the test did not measure blind discovery or exploit validation.
Key points
- Aikido introduced Altar, an open-weight model intended to run within customer infrastructure, including air-gapped environments, through its Aikido Machine appliance.
- The company says pruning and AWQ quantization reduced GLM-5.3 from 1,506.7 GB to 328 GB; Altar retains 168 of 256 routed experts.
- In Aikido’s internal benchmark of 32 known vulnerabilities across 30 repositories, Altar averaged 60.4% recall per run and rediscovered 23 vulnerabilities at least once across three runs.
- The quantized GLM-5.3 baseline averaged 61.5% recall and also covered 23 vulnerabilities; the full-precision model averaged 65.6% and covered 25.
- Aikido says the benchmark measured targeted vulnerability rediscovery, not blind discovery across codebases, exploit validation, or fix proposals.
- The company says Altar identified a critical-severity vulnerability during a client production pentest after deployment, but provides no further details.
Article Details
- Event Type
- Open-weight AI security model launch
- Impact
- Aikido introduced Altar, a compressed open-weight model intended to support security work within customer-controlled, including air-gapped, infrastructure. Aikido reports that Altar is 328 GB—78.2% smaller than the full-precision GLM-5.3 and 32.8% smaller than its quantized parent—and retained 23 of 25 vulnerabilities covered by the full-precision model in its internal benchmark. Aikido also says Altar identified a valid critical-severity vulnerability during a client production pentest after deployment. The benchmark measured targeted CVE rediscovery, not blind discovery across a codebase, exploit validation, or fix proposals.
People
Vendors
AikidoWe took our first step toward sovereign security intelligence with our autonomous pentesting appliance Aikido Machine, which runs entirely inside a customer’s own infrastructure, including fully air-gapped environments.Cerebrasfootprint of LLMs, so we decided to go with a technique that can preserve performance for our use cases: Cerebras REAP, Router-weighted Expert Activation Pruning. We used REAP’s contribution score with aZ.aiThanks to Z.AI for GLM-5.3, Cerebras for REAP, cyanwiki for the quantized model, and 0xSero for the compression work. The public fidelity study and keep plans and pruning toolkit provide more technical detail, without
Products
Aikido MachineWe took our first step toward sovereign security intelligence with our autonomous pentesting appliance Aikido Machine, which runs entirely inside a customer’s own infrastructure, including fully air-gapped environments.AltarToday, we’re introducing Altar, our first open-weight security model, built to bring frontier-grade defensive security into infrastructure you control.GLM-5.3We started with GLM-5.3, one of the strongest models in our security evaluations. The full model occupies 1.51 TB; quantization brings that down to 488 GB, and our expert-pruning process reduces it further to 328 GB,