GPT-5.5-Cyber
2026-05-07 · Publisher : OpenAI · Domain : Artificial Intelligence
OpenAI is rolling out GPT-5.5-Cyber in limited preview for critical-infrastructure defenders, a permissive variant of GPT-5.5 for penetration testing and authorized exploit validation.
What changes
- Most permissive behavior in the lineup, with stronger verification and account-level controls.
- Target use cases: red teaming, penetration testing, controlled exploitability validation.
- No raw capability gain expected over GPT-5.5: the preview does not beat GPT-5.5 on every cyber evaluation.
- For most teams, GPT-5.5 with Trusted Access for Cyber remains the recommended starting point.
- Individual members must enable Advanced Account Security starting June 1, 2026.
### Specifications - Based on GPT-5.5 (see its page for context and modalities). - Availability: limited preview for verified defenders, no public access. - Completion rate on advanced cyber requests: 57.3% for GPT-5.5-Cyber, a figure published later in the August Daybreak post. - API identifier: not disclosed. ### Pricing Pricing not disclosed at launch. ### Benchmarks No numerical score published by the vendor in the announcement. ### Key takeaways Second model in the Cyber series, with no public price premium and no dedicated benchmark. Its 57.3% completion rate was judged insufficient by researchers, which motivated GPT-5.6-Cyber.
Links
Tags : LLM, OpenAI, GPT-5, Cybersécurité, Accès restreint