OpenAI has introduced three new artificial intelligence models to its GPT-5.6 family, following a commercial release delay prompted by White House security reviews.
The San Francisco-based technology company stated that the new software releases, GPT-5.6 Sol, GPT-5.6 Terra and GPT-5.6 Luna, outperform rival market systems Claude Mythos 5 and Fable 5 across a variety of performance benchmarks.
The widespread product rollout was originally scheduled to take place in June. However, the official launch was postponed after the Donald Trump administration demanded that OpenAI first grant early system access to U.S. government-approved organisations, according to a report published by The Information.
Sources familiar with the decision linked the government mandate directly to White House anxieties regarding national security and technology safety. Anthropic, the developer of the competing Claude models, reportedly faced an identical regulatory requirement prior to its own software deployment.
Flagship options and free models
The flagship model, GPT-5.6 Sol, is restricted to a paid subscription tier and is engineered for highly complex computing operations. OpenAI tailored this system specifically for software development, cybersecurity configuration, deep research and big data processing. Sol also includes a specialised «Ultra» mode that allows users to integrate multiple additional autonomous digital agents within a single workflow.
The second model, GPT-5.6 Terra, serves as the company’s new free variant. The firm claims Terra delivers operational efficiency and performance on par with the Fable 5 architecture. The final model, GPT-5.6 Luna, targets everyday personal productivity, focusing on standard web information search and workplace workflow automation.
In addition to individual tier adjustments, all three models feature upgraded functionality for analysing documents, building professional presentations and creating digital spreadsheets.
Officials at OpenAI stated that development teams prioritised safety protocols during the training phase. The newly released systems contain updated internal guardrails designed to actively restrict users from exploiting the artificial intelligence for malicious activities.
