| Entity | OpenAI |
|---|---|
| Counterparty | ChatGPT users, free and paid plans, globally |
| Date | 7 Oct 2026 |
| Instrument | GPT-6 Sol and GPT-6 Luna: October 2026 update (Deployment Safety Hub system card) |
| Region | U.S. & Allies |
OpenAI is launching GPT-6 to everyone in ChatGPT, across free and paid plans globally, and these models will replace GPT-5.6 Sol and GPT-5.6 Luna in ChatGPT.
The 7 October 2026 system card treats this October release of GPT-6 Sol and GPT-6 Luna as High capability in Cybersecurity and in Biological and Chemical under the Preparedness Framework, and states that neither reaches the High threshold in AI Self-Improvement. Codex and ChatGPT Work stay on the September versions.
Who this affects
- For investors: The card states GPT-6 Sol performed comparably to GPT-5.6 Sol on automated cyber evaluations without a clear improvement in capabilities, and that GPT-6 Luna underperformed GPT-6 Sol; on SEC-Bench Pro the scores are 68.85% (Sol) and 48.50% (Luna), against 85.4% for GPT-6 Astra.
- For counsel/compliance: Safeguards for GPT-6 Sol (October) and GPT-6 Luna (October) are the same set detailed in the GPT-5.6 System Card; October is the ChatGPT cut, and September remains in Codex and ChatGPT Work.
- For policymakers: The Preparedness treatment is High in Cybersecurity and Biological and Chemical, below Critical in cybersecurity, and below High in AI Self-Improvement; the Safety Advisory Group recommended both models were below the critical cyber threshold, and OpenAI leadership determined the safeguards sufficient for public launch.
The read
The October cut is a ChatGPT replacement, not a cyber step-up on the evaluations in this card. Sol is described as comparable to GPT-5.6 Sol without a clear improvement; Luna underperforms Sol; both are held below Critical, and the safeguard stack is the GPT-5.6 set.
The High Biological and Chemical treatment is not the same as clearing every listed High-threshold metric: Sol is at 25.62% on ProtocolQA Open-Ended against a 54% High threshold and 72.45% on Tacit Knowledge and Troubleshooting against 80%, while Luna is at 94.00% on that tacit-knowledge eval and 34.57% on ProtocolQA.
Reported Critical bio results do not cross the indicative Critical thresholds, and separate Critical testing was not required for Luna because it scored below GPT-5.6 Sol on all High capability evaluations.
What was the action?
OpenAI launched GPT-6 Sol (October) and GPT-6 Luna (October) to everyone in ChatGPT, free and paid, globally, replacing GPT-5.6 Sol and GPT-5.6 Luna in ChatGPT.
Users accessing GPT-6 Sol and GPT-6 Luna in Codex, and via ChatGPT Work, are still on previously released versions; the card labels today’s cut October and the Codex and Work versions September. GPT-6 in ChatGPT incorporates Astra’s safety advances, with safety training updated against high-risk misuse in cyber, biology, and violence.
What’s the penalty / scope, and how is it set?
No penalty. The action is a public deployment with a Preparedness determination. High cybersecurity is defined in the card as a model that “removes existing bottlenecks to scaling cyber operations including by automating end-to-end cyber operations against reasonably hardened targets or by automating the discovery and exploitation of operationally relevant vulnerabilities.” Both models are treated as High and below Critical in cybersecurity.
At maximum reasoning effort, ExploitBench is 82.62% for Sol and 44.66% for Luna; the card notes those results may be inflated by exposure to historical vulnerabilities in the public benchmark. On Biological and Chemical High-threshold evals, Multimodal Troubleshooting Virology is 51.68% (Sol) and 48.13% (Luna) against 31%; TroubleshootingBench is 41.83% and 41.46% against 36.4%. Critical bio figures for Sol — Mean R² 0.23 against 0.60, composite 0.349 against 0.75, negative log-likelihood 12.95 against ≤ 9.40 — did not cross the indicative Critical thresholds. Safety evals are run at the lowest reasoning deployment settings; capability assessments at maximum reasoning effort.
Who else is exposed?
ChatGPT users on free and paid plans globally receive the October models. Codex and ChatGPT Work users remain on the September versions. The card does not name other companies or sectors. On production benchmarks the card reports a regression for Sol on standard self-harm, and for Luna on standard self-harm, gore, and sexual content; manual review found the violative responses borderline and generally low severity, and the evals do not include system-level interventions such as Trusted Contact, localized crisis helplines, and Parental Controls.
What happens next
The card states OpenAI continues monitoring these categories after launch to evaluate online performance and further adjust safeguards as appropriate. It does not set a date for moving Codex or ChatGPT Work off the September versions. The internal Safeguards Report informed the Safety Advisory Group’s recommendation and leadership’s determination that the safeguards are sufficient for this public launch; additional detail in that report is withheld as potentially useful to attackers.
"Under our Preparedness Framework, we are treating this October release of GPT-6 Sol and GPT-6 Luna as High capability in both Cybersecurity and Biological and Chemical domains."OpenAI Deployment Safety Hub, GPT-6 October 2026 →