OpenAI launches GPT-5.6 amid regulatory challenges
What's the story
OpenAI is rolling out its latest AI model suite, GPT-5.6, to a small group of trusted partners. This comes just a day after the Trump administration asked the firm to stagger its release. The new suite includes three models: Sol (the flagship), Terra (a medium-tier model for "high-volume work"), and Luna (an everyday model that is fast and affordable). The company claims these models excel in coding, cybersecurity, biology tasks, and long-horizon agentic AI tasks.
Cost details
New models are cheaper than competitors
The pricing for the new models is significantly lower than that of their competitors. For instance, per million tokens, GPT-5.6 Sol costs $5 input/$30 output, almost half the price of Anthropic's Claude Fable 5 ($10 input/$50 output). Terra costs half as much as Sol, while Luna is less than half the price of Terra. OpenAI has also introduced two new modes for Sol: a "max" mode for deeper reasoning and an "ultra" mode to leverage sub-agents.
Safety measures
Focus on safety and cybersecurity
In light of recent security concerns, OpenAI has focused heavily on safety in its latest model. The company says GPT-5.6 is trained to refuse prohibited cyber assistance, including when users attempt to disguise their intent or jailbreak the model. It also claims that Sol is better at helping people find and fix vulnerabilities than reliably carrying out end-to-end attacks.
Enhanced protections
Extra-sensitive approach during the preview period
OpenAI has also added stronger protections for higher-risk activity, sensitive cyber requests, and repeated misuse in Sol. The company has dedicated around 700,000 A100e GPU hours to automated red-teaming and worked with third-party testers. These tests will continue over the next two weeks as part of an extra-sensitive approach during the preview period, closely monitored by the Trump administration.
Access concerns
OpenAI hopes government access process isn't long-term default
OpenAI has expressed its belief that the government access process shouldn't become a long-term default. The company argues it keeps the best tools from users, developers, enterprises, cyber defenders, and global partners who need them. Despite cooperating with the US government ahead of this launch, OpenAI hopes this isn't a trend for future model releases.