OpenAI’s GPT-5.6, branded as Sol, finally launches Thursday after the U.S. government stalled its release. The Department of Commerce approved the public rollout only after the Center for AI Standards and Innovation conducted extra safety tests. OpenAI criticized the hold, arguing it kept cutting-edge tools from developers and businesses. The company updated its benchmarks: Sol scores 88.8% on TerminalBench 2.1, while the top-tier Sol Ultra hits 91.9%, edging past Anthropic’s Claude Mythos 5 at 88%. Sol also matches Mythos 5 on cybersecurity tasks, but uses a third fewer tokens. Pricing is lower too — $5/$30 per million tokens versus Anthropic’s Fable 5 at $10/$50. Why it matters: This showdown highlights the growing tension between AI innovation and government oversight, with no binding standards yet in place — leaving the industry in a regulatory gray zone where speed and safety keep butting heads.