OpenAI Launches GPT-5.6 Suite Under White House Oversight

The flagship model, Sol, enters the market at $5 per million tokens for input and $30 for output, undercutting rival Anthropic’s Claude Fable 5. OpenAI has positioned the suite as a specialist in coding, cybersecurity, and biology, while introducing "max" and "ultra" modes that allow for deeper reasoning and the deployment of sub-agents. The latter feature hints at the influence of Peter Steinberger, whose work at the company appears to be gaining traction.

Faced with a security-focused climate in Washington, D.C., the company devoted significant resources to detailing its safety protocols. OpenAI claims to have utilized roughly 700,000 A100e GPU hours for automated red-teaming to ensure the model resists jailbreaking attempts and cyber-offensive misuse. Despite these measures, the company acknowledged that safeguards might occasionally block legitimate work, a friction point that will be monitored during a preview period requiring case-by-case approval from federal officials.

OpenAI expressed a desire to move past these restrictions, stating that such government-led access processes should not become a long-term norm. The company aims for general availability in the coming weeks, viewing the current setup as a temporary bridge while it negotiates the specifics of a future executive order framework for model releases.