Breaking

OpenAI Debuts GPT-5.6 Sol in Limited Preview, Tuned for Cybersecurity

June 26, 2026 at 13:28 EDT

Foundation Models
Security
AI Agents

OpenAI on June 26, 2026 launched a limited preview of its new GPT-5.6 family (Sol, Terra and Luna), positioning the top-end Sol as its most capable model yet for cybersecurity. Access is restricted to trusted partners via the API and Codex, with no ChatGPT preview for now and a broader release expected in a few weeks (OpenAI, help.openai.com).

June 2026 · OpenAI

GPT-5.6 "Sol" Debuts as a Government-Gated Preview

OpenAI's most capable model yet launches to trusted partners only — under per-customer approval requested by the US government over cybersecurity concerns. General availability is expected within weeks.

Sol

Maximum capability flagship — agentic workloads, max reasoning + ultra mode.

Terra

GPT-5.5-level performance at ~2× the cost efficiency.

Luna

Fast, low-cost model for high-volume workloads.

CyberGym benchmark

GPT-5.5-Cyber sets a new SOTA score

81.8%

GPT-5.5

85.6%

GPT-5.5-Cyber

Tiered access by use case & verification

Default — GPT-5.5

General developers · standard guardrails

Trusted Access for Cyber (TAC)

Verified defenders · binary reverse engineering, etc.

GPT-5.5-Cyber (limited)

Red teaming / penetration testing · most permissive

3,000+

critical vulnerabilities automatically fixed via the Codex Security plugin.

20+

vendors in the Cyber Partner Program around the Daybreak platform.

Praised

Shift from vulnerability discovery to patch automation
Binary reverse engineering without source code
Expected edge in coding, workflows, bio/cyber

Criticized

Government-imposed release restrictions
Dual-use risk serving both defense and attack
Sol specs largely undisclosed; user experience limited

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.