Models
How Sebastion AI picks an audit model based on your plan.
Models
Sebastion AI picks the audit model for you based on your plan. There
is no model parameter to set. We pick a frontier model that gets
security review right and route around it when something better lands.
| Plan | What we run today |
|---|---|
| Free | A fast frontier model (currently Claude Sonnet 4.6) |
| Pro | A higher-reasoning frontier model (currently Claude Opus 4.6) |
The exact mix can change with provider availability and pricing. When it changes materially, we'll publish a note on the blog and notify customers directly.
Why we don't expose model selection
Security review is the product. Choosing the model is our job, not yours: we benchmark new releases against our regression suite, route around outages and roll back models that regress on false-positive rates. Letting you pick a model would mean letting you pick a worse audit, so we don't.
If you have a strong opinion on model choice (e.g. data-residency constraints in a specific region), reach out via contact and we'll see what we can do.
If a new model genuinely beats the current one on our regression suite, we switch. You will find out via your audits getting slightly better, not via a newsletter.
.sebastionai.yml config
Per-repository configuration for severity thresholds, ignored paths, disabled rules and scanner toggles.
False positives and Learnings
How Sebastion AI suppresses noisy findings — chat commands, AGENTS.md auto-detect, and operator overrides — with the security properties that keep suppression safe.