wrath labs est. 2026

we build models that dominate cyber

Frontier large language models trained end-to-end for offensive and defensive cyber operations. No hands holding. No restrictive controls. Just raw capability.

scroll ↓

what we're doing

Most labs are building models to talk to people. We're building models to act on networks — to find the way in, understand what they've found, and decide what to do next, without a human narrating every step.

Cyber operations are a narrow domain with a wide surface: reconnaissance, exploit development, malware research, lateral movement, persistence, cleanup. We think a model purpose-built for that domain, rather than adapted to it, ends up somewhere general-purpose systems don't reach.

the models

wrath:1b
Edge deployments — optional reasoning, agentic capabilities, tuned for fast latency and high tok/s
wrath:9b
Vision-capable — optional reasoning, agentic capabilities, medium latency, more accurate responses
wrath:120b
Flagship, coming soon — multimodal, optional reasoning, agentic capabilities, MoE architecture

the work

focus
Offensive & defensive tooling for autonomous cyber operations — red team automation, exploit development, TTP coverage, agentic red teaming with 0 refusals
stage
Pre-launch — training and optimizing our models for maximum dominance
team
Small, technical, heads-down team with a proven track-record

join the waitlist for hosted inference access

By entering your email, you agree to receive updates from wrath labs. See our privacy policy.