Open-weight models for security work.
ZeroDayAPI builds on open-weight base models and tunes them to engage with authorized security tasks instead of refusing them.
MiniMax M3.1 Flash Abliterated
Our next model for exploit analysis, reversing and triage. Early access is by request for company accounts.
Request early accessDeveloper demo IDs
Three model IDs return demo responses so you can build against the request shape today. Production inference is still being connected.
See the IDs in the docsDemo model IDs
| Model ID | Focus | Status |
|---|---|---|
zeroday-offensive-70b | API pen testing, exploit analysis | Demo |
zeroday-reverse-32b | Assembly reversing, patch diffing | Demo |
zeroday-triage-8b | Security telemetry triage | Demo |
What “abliterated” means
Abliteration is a technique that suppresses a model’s refusal behaviour, so it answers questions about exploits, malware and payloads that general-purpose assistants decline. It is not a licence to misuse: access is limited to authorized work under our Acceptable Use Policy.
Licences and base models
Each model starts from an open-weight base model and keeps that model’s licence terms. The base model and its licence will be listed with every model at launch. Third-party logos on the home page show the wider open-weight ecosystem and do not imply endorsement.
Evaluation
We do not publish performance or refusal-rate scores until the evaluation has been run. The tasks we plan to measure are on the benchmarks page.
Questions about a model?
Write to [email protected].