GPT-5.5-Cyber logo

GPT-5.5-Cyber

OpenAI · Released Jun 2026

Conditional

85.6% CyberGym — top single-model score — but restricted-access for vetted defenders only.

Is it right for you?

Good for

  • Vetted cybersecurity professionals performing vuln triage, malware analysis, penetration testing, and secure code review at scale.

Not good for

  • General-purpose coding or non-security tasks — standard GPT-5.5 is better suited and widely available.

How it performs by task

Vulnerability discovery and reproduction

Excellent

85.6% CyberGym — highest single-model score.

Exploit development (defensive)

Very Good

39.5% ExploitGym, up from 25.95% for standard GPT-5.5.

Secure code review

Excellent

Codex Security scanned 30M+ commits; 70,000+ fixes confirmed.

General reasoning / non-security coding

Good

69.8% SEC Bench Pro. Decent but not its design focus.

Pricing

Input

N/A

Output

N/A

Context

Trusted Access for Cyber; not commercial API

View full pricing

Benchmarks

BenchmarkScoreSource
CyberGym85.6% Source
ExploitGym39.5% Source
SEC Bench Pro69.8% Source

No verdict changes yet

The clock starts day one — changes land here as our verdict evolves.

Verification log

  • Pricing— No changes

    Automated agent

  • Pricing— No changes

    Automated agent

  • Pricing— No changes

    Automated agent

  • Pricing— No changes

    Automated agent

    Pricing unchanged: N/A. Restricted-access model — gpt-5.4-cyber listed with no pricing on OpenAI API pricing page. Trusted Access for Cyber program only.

  • Pricing— No changes

    Automated agent

    Pricing remains N/A — restricted access for vetted defenders. Codex Security: 30M+ commits scanned, 70,000+ fixes confirmed.

  • Pricing— No changes

    Automated agent

  • Pricing— No changes

    Imported at launch

  • Profile— No changes

    Imported at launch

How we evaluate