Audn AI releases unrestricted Kimi K3.1 Pro model for offensive cybersecurity

Audn AI releases an unrestricted Kimi K3.1 Pro model for offensive cybersecurity, with weights on Hugging Face and access through several paid and free routes.

The company says the model scores about 33% on its refusal benchmark, compared with about 77% for GLM5.3 and about 80% for DeepSeek, and that it is still training the weights on cybersecurity data.

Access options, as the company lists them: running the Hugging Face weights on eight B300 GPUs, which the company estimates at about $60K; a $999 Cyber program on audn.ai/necromicon with unlimited use capped at three parallel sessions; pay-as-you-go access at platform.audn.ai with no KYC and a $5 minimum top-up; and one day of free access at penclaw.ai, which requires KYC. Audn AI says some users spent $30K in tokens on the $999 program. Some of its sites still show DeepSeek because the company is moving them to Kimi K3.1 Pro.