Cantina open-sources a 66.7% security model

Cantina Security has released Apex Flash-1, an open-weights model that its Hugging Face model card says solved 40 of 60 held-out security tasks (66.7%) at an estimated $2.38 for the run. The same table puts GLM-5.3-Flash at 36/60 and Claude Opus 5 High at 43/60, but the evaluation is company-run and limited to 20 vulnerability cases; the model card says the standard checkpoint’s result does not apply to its modified-refusal derivative. Security teams should treat the release as a cheap, local research candidate—not a validated autonomous defender—and compare it against an authorized test set, as GLM-5.3-Flash’s price-performance tradeoff shows why cost alone is not a deployment verdict.