# GPT-5.5-Cyber

OpenAI, released 2026-06-22

**Verdict: conditional.** 85.6% CyberGym — top single-model score — but restricted-access for vetted defenders only. Last verified 2026-06-24.

## Pricing

- Input: N/A
- Output: N/A
- Context: Trusted Access for Cyber; not commercial API

## Strengths

- Vetted cybersecurity professionals performing vuln triage, malware analysis, penetration testing, and secure code review at scale.
- General-purpose coding or non-security tasks — standard GPT-5.5 is better suited and widely available.

## Benchmarks

- CyberGym: 85.6% (https://openai.com/index/patch-the-planet/)
- ExploitGym: 39.5% (https://cryptobriefing.com/openai-patch-planet-open-source-security/)
- SEC Bench Pro: 69.8% (https://cryptobriefing.com/openai-patch-planet-open-source-security/)

## Sources

- [OpenAI — Patch the Planet](https://openai.com/index/patch-the-planet/)
- [Axios — GPT-5.5-Cyber more capable version](https://www.axios.com/2026/06/22/openai-rolls-out-more-capable-version-of-cyber-model)
- [Crypto Briefing — Patch the Planet details](https://cryptobriefing.com/openai-patch-planet-open-source-security/)

Canonical: https://neomanex.com/models/gpt-5-5-cyber