OpenAI has launched its new AI model generation, “GPT-5.6,” which has been publicly available since July 9. The model introduces a new naming scheme using the permanent performance classes Sol, Terra, and Luna, representing first, second, and economy class tiers. Free and Go plan users are assigned the mid-tier Terra model by default, while Plus, Pro, Business, or Enterprise subscribers can switch between all three models. A new “Ultra” mode, where multiple AI agents work simultaneously, is reserved for Pro and Enterprise customers. New features include the productivity tool “ChatGPT Work,” which accesses connected apps and files to create documents, spreadsheets, or presentations in desktop apps for Mac and Windows, and the integration of the programming tool “Codex” into the ChatGPT app. Since June, an executive order by U.S. President Trump has required developers to submit their most powerful models to the government for review prior to release, leading GPT-5.6 to be initially available only to select partners due to its cybersecurity capabilities. OpenAI calls GPT-5.6 its most powerful cybersecurity model to date; in tests, it identified vulnerabilities in the Chrome and Firefox browsers but was unable to construct a complete, functional attack. On an independent index measuring AI programming performance, OpenAI claims a new top score just above Anthropic’s “Claude Fable 5” model.