Bottom line: OpenAI introduces GPT-5.6 Sol, specifically optimized for vulnerability detection and achieving competitive performance with significantly fewer tokens.
OpenAI has announced a limited preview of GPT-5.6, with the flagship model Sol specifically designed for analyzing software vulnerabilities and complex logical tasks. The release initially occurs only for selected partners following consultation with the U.S. government.
OpenAI divides the new GPT-5.6 model generation into three performance tiers: Sol as the most powerful model, Terra for everyday tasks at lower cost, and Luna as a fast, cost-effective variant. The version number henceforth designates the model generation, while the names describe the permanent capability tiers.
Sol demonstrates comparable results to competing models (such as Mythos Preview) in cybersecurity tests on ExploitBench, but requires only one-third of the output tokens. When analyzing source code from Chromium and Firefox, the model could isolate software errors and fundamental mechanisms, but did not create functioning complete attack chains. OpenAI emphasizes that Sol is better suited for defensive tasks such as vulnerability detection and patch development than for offensive scenarios.
The infrastructure features multi-layered security barriers: automated real-time classifiers examine requests from biology and cybersecurity domains, with generation paused upon suspicious input and a secondary model verifying context. OpenAI invested over 700,000 GPU hours in automated security testing to counter universal system bypasses. Limited availability serves as a temporary measure to assess national security risks in accordance with current regulations.
Broader deployment to ChatGPT and the public API is planned for the coming weeks.
Source: www.it-daily.net · Published June 30, 2026
Lumi AI News — AI-assisted curation pursuant to Article 50 EU AI Act. Paraphrase and classification through Lumi News Pipeline v1.7.2.