OpenAI has released the GPT-5.6 model family for general availability, introducing three tiers: Sol (flagship), Terra (balanced everyday model), and Luna (cost-efficient), alongside ChatGPT Work, a new agentic assistant designed for enterprise task automation.
The launch pairs significant capability gains with what OpenAI calls its “most robust safeguards to date,” specifically designed to target cybersecurity misuse.
GPT-5.6 Sol achieves state-of-the-art results across coding, cybersecurity, and scientific reasoning while using fewer tokens than competing models.
OpenAI Launches GPT-5.6 and ChatGPT Work
On Agents’ Last Exam, a benchmark spanning 55 professional fields, Sol scored 53.6, 13.1 points ahead of Claude Fable 5, while smaller tiers Terra and Luna matched or exceeded Fable 5 at roughly one-sixteenth the cost.
A new “ultra” mode coordinates multiple agents in parallel for demanding, long-horizon tasks. The most notable technical shift is GPT-5.6’s cybersecurity performance.
On ExploitBench2, which tracks the progression from vulnerable code to arbitrary code execution, Sol scored 73.5% compared to GPT-5.5’s 47.9%.
On ExploitGym3, which tests real-world vulnerability-to-exploit conversion, Sol nearly doubled the previous model’s pass rate (24.9% vs. 15.1% under a two-hour cap, reaching 33.7% at six hours). SEC-Bench Pro results showed similar gains, with Sol hitting 71.2% against GPT-5.5’s 45.8%.
OpenAI states this increased capability supports defensive work, such as secure code review, patch validation, threat modeling, and detection engineering, rather than crossing into what it defines as “Critical” risk territory.
The company’s testing suggests GPT-5.6 is more capable at finding and fixing vulnerabilities than executing autonomous end-to-end attacks against hardened targets.
To manage dual-use risk, OpenAI built a multi-layered safety system combining trained-in model protections, real-time reasoning-based monitoring, and account-level access controls.
Compared to previous models, GPT-5.6 Sol’s cyber safeguards block roughly 10 times as much potentially harmful activity, according to OpenAI.
Access to the model’s fullest defensive capabilities is now gated behind a “Trusted Access for Cyber” program under OpenAI Daybreak.
Verified individuals and organizations working in authorized environments such as vulnerability triage, malware analysis, and detection engineering get expanded access.
Notably, individual users must enable Advanced Account Security with hardware-backed passkeys by September 1, 2026, to retain access to OpenAI’s most cyber-capable frontier models; those who don’t will be reverted to default, more restricted access.
OpenAI is also partnering with Yubico to offer discounted hardware keys. Pre-launch validation included approximately 700,000 A100e GPU-hours of automated red teaming, plus external expert testing, feeding into an updated GPT-5.6 system card.
Alongside the model launch, OpenAI introduced ChatGPT Work, an agent capable of multi-hour autonomous workflows across connected apps (Slack, Notion, Microsoft 365, Google Drive).
For security teams, this raises governance considerations: admins can manage plugin access, restrict sensitive actions, and use an “Auto-review” layer that inspects high-risk actions before execution. The Compliance API extends visibility into ChatGPT Work conversations for enterprise oversight.
Follow us on Google News , LinkedIn and X to Get More Instant Updates. Set Cyberpress as a Preferred Source in Google.
