Bottom line: GPT-5.6 Sol achieves comparable results to competing models in cybersecurity tests with only one-third of the token output and is primarily designed for defensive tasks such as vulnerability analysis.
OpenAI has announced a limited preview of GPT-5.6, a new model generation available in three performance tiers. The flagship model Sol is specifically tailored for complex logical tasks and software vulnerability analysis.
OpenAI has introduced GPT-5.6 as a new model generation, divided into three performance tiers: Sol for complex logical tasks, Terra for everyday tasks at reduced costs, and Luna as a fast, cost-effective variant. The version number will henceforth designate the model generation, while the names describe persistent capability levels. The release initially takes the form of a limited preview for selected partners following consultation with the US government.
In security tests on the ExploitBench platform, Sol achieved comparable results to competing models such as Mythos Preview, but required only one-third of the output tokens. When analyzing source code repositories such as Chromium and Firefox, the model identified software defects and explored security mechanisms, but did not create functional attack chains. OpenAI emphasizes that Sol is designed for defensive tasks – particularly for identifying vulnerabilities and developing patches.
To mitigate risk, OpenAI employs a multi-layered security architecture. Automated real-time classifiers review inputs in the biology and cybersecurity domains, pause suspicious generations, and route them to a secondary model for context validation. Ahead of release, over 700,000 GPU hours were spent on automated security testing against universal system bypasses. Broader availability through ChatGPT and the API is planned for the coming weeks.
Source: www.it-daily.net · Published June 30, 2026
Lumi AI News — AI-assisted curation pursuant to Art. 50 EU AI Act. Paraphrase and classification via Lumi News Pipeline v1.7.2.