OpenAI Released GPT-5.6 Sol With Limited Access and Strong Cyberattack Protections

HomeCyber Security OpenAI Released GPT-5.6 Sol With Limited Access and Strong Cyberattack Protections By Guru Baran June 29, 2026 OpenAI has officially begun a limited preview of the GPT‑5.6 model series Sol, Terra, and Luna, positioning its flagship Sol as the company’s most capable and security-hardened AI model to date, available initially only to a small group of trusted partners at the formal request of the Trump administration. The GPT‑5.6 family introduces three distinct capability tiers under a new naming system. Sol is the flagship model; Terra is a balanced model for everyday work, delivering competitive performance to GPT‑5.5 at 2x lower cost; and Luna is a fast, affordable model designed to bring strong AI capability at the lowest price point in the lineup. In the new naming convention, the number identifies the model’s generation, while Sol, Terra, and Luna identify durable capability tiers that can each advance on their own cadence. Pricing for GPT‑5.6 is structured per million tokens: Sol at $5 input / $30 output, Terra at $2.50 input / $15 output, and Luna at $1 input / $6 output. OpenAI is also launching GPT‑5.6 Sol on Cerebras at up to 750 tokens per second in July, bringing unprecedented inference speed to select enterprise customers. The controlled rollout of GPT‑5.6 comes after the Trump administration formally requested that OpenAI stagger the public release, citing the model’s advanced capabilities and national security implications. According to reporting, the request came from the Office of the National Cyber Director and the Office of Science and Technology Policy, with Commerce Secretary Howard Lutnick also reportedly advising against an unrestricted launch. OpenAI CEO Sam Altman informed employees that the government would be “approving access customer by customer” during the preview period. OpenAI has been explicit, however, that this arrangement is not its preferred long-term model, stating: “We don’t believe this kind of government access process should become the long-term default. It keeps the best tools from users, developers, enterprises, cyber defenders, and global partners who need them.” During the preview period, GPT‑5.6 models will be accessible via the API and Codex to select trusted partners, with broader availability across ChatGPT, Codex, and the API planned for the coming weeks. GPT-5.6 Sol’s Cybersecurity Capabilities GPT‑5.6 Sol is OpenAI’s most capable model for long-horizon security tasks, including vulnerability research and exploitation, significantly shifting the performance-efficiency frontier. On ExploitBench, Sol is competitive with Anthropic’s Mythos Preview while using only approximately one-third of the output tokens. On ExploitGym, a benchmark developed by UC Berkeley researchers in collaboration with OpenAI and other frontier labs, Sol, Terra, and Luna all demonstrate significant improvements in cyber capabilities as reasoning effort increases. For coding workflows, GPT‑5.6 Sol sets a new state of the art on Terminal-Bench 2.1, scoring 88.8%, with the Sol Ultra configuration pushing the benchmark to 91.9%, outperforming Claude Mythos 5 (84.3%) and GPT‑5.5 (88.0%). The model also introduces a new ultra mode that leverages subagents to accelerate complex multi-step work beyond the capability of any single agent. Layered Cyberattack Protections GPT‑5.6 Sol launches with OpenAI’s most robust safety stack to date, specifically designed to withstand real-world adversarial pressure. The safety architecture is multi-layered: Model-level safeguards trained to refuse prohibited cyber assistance, including jailbreak attempts Real-time misuse classifiers that evaluate output as it is generated, with the ability to pause generation and route flagged content to a larger reasoning model for review Account-level review that analyzes signals across conversations to distinguish persistent malicious behavior from legitimate dual-use security research Differentiated access controls that preserve access for legitimate defensive work without making the most sensitive capabilities broadly available Automated red-teaming using over 700,000 A100-equivalent GPU hours dedicated to finding universal jailbreaks and hardening the safeguard stack Human expert red-teaming by third-party testers, continuing through the preview period Critically, GPT‑5.6 Sol does not cross the Cyber Critical threshold under OpenAI’s Preparedness Framework. In evaluations involving Chromium and Firefox, it identified bugs and exploitation primitives but did not autonomously produce a functional full-chain exploit under tested conditions. OpenAI’s stated goal is to make “prohibited offensive activity more difficult, uncertain, and detectable” without limiting legitimate use cases such as code review, vulnerability research, patch development, and defensive security testing. GPT‑5.6 introduces a max reasoning effort mode, allowing Sol the maximum time to reason deeply before generating output. The new ultra mode goes beyond a single-agent architecture, leveraging subagents in parallel to accelerate complex, long-horizon tasks. On GeneBench v1, which evaluates long-horizon genomics and quantitative biology analyses, Sol achieves stronger results than GPT‑5.5 while using fewer tokens, demonstrating efficiency gains alongside raw capability improvements. Prior to the June 26, 2026, launch, The Information first reported that OpenAI had agreed to stagger the GPT‑5.6 release after the Trump administration formally requested controlled access, citing national security concerns. The move followed a high-profile incident on June 12, 2026, when the administration issued export control directives compelling Anthropic to take its Fable 5 and Mythos 5 models entirely offline to prevent access by foreign nationals, setting a critical precedent for frontier AI governance. Shortly before the GPT‑5.6 series launch, OpenAI officially released GPT‑5.5‑Cyber, a specialized model engineered for automated vulnerability detection, patch generation, and remediation at machine speed, as part of its broader Daybreak initiative. The model delivered state-of-the-art results across three major benchmarks: CyberGym (85.6%), ExploitGym (39.5%), and SEC-bench Pro (69.8%) all surpassing GPT‑5.5 baselines. What Features Should AI SOC Have? – Download Free 2026 AI SOC Features Checklist Tags cyber security cyber security news Copy URL Linkedin Twitter ReddIt Telegram Guru Baranhttps://cybersecuritynews.com Gurubaran KS is a cybersecurity analyst, and Journalist with a strong focus on emerging threats and digital defense strategies. He is the Co-Founder and Editor-in-Chief of Cyber Security News, where he leads editorial coverage on global cybersecurity developments. Trending News White House Orders Federal Agencies to Migrate Systems to Post-Quantum Cryptography Microsoft Entra Conditional Access Policies Can Be Bypassed Via Nested App Authentication Chinese Cyber Contractors Use Malware, Botnets, and Stolen Data to Enable State Operations Your SOC Has Too Many IOCs: How to Cut Feed Noise, Prioritize What Matters, and Improve Response  EvilTokens Hides Its Attack Flow in the Browser, Exposing Static Analysis Gaps   Latest News Cyber Security New Bucket Hijacking Attack Allows Hackers to Reroute Cloud Data Streams to External Storage Cyber Security New DirtyClone Linux Vulnerability Allows Attackers to Gain Root Access Via Cloned Packets Cyber Security Amazon Q Vulnerability Let Attackers Execute Code and Access Sensitive Cloud Environments Cyber Security New Linux pedit COW Exploit Allows Attackers to Gain System Root Access Technology Why 24/7 Support Is Essential for Cybersecurity Providers