OpenAI announced GPT-5.6, its latest AI model family, just two months after GPT-5.5 rolled out. The company is taking a different approach this time with a staggered release involving U.S. government coordination. Three distinct models debut under a new celestial naming scheme: Sol as the flagship, Terra for balanced performance, and Luna as the fast, affordable option.

The new naming strategy separates generation numbers from capability tiers, giving developers clearer choices across intelligence, speed, and cost. GPT-5.6 introduces enhanced reasoning with 'max' and 'ultra' modes for Sol, plus subagent support for complex tasks. OpenAI claims improvements in coding, biology, and cybersecurity capabilities, though access remains restricted to select partners during the preview phase.

Pricing spans $1 to $30 per million tokens depending on model selection. Sol costs $5 input/$30 output, Terra runs $2.50 input/$15 output, and Luna sits at $1 input/$6 output. The models also feature improved prompt caching with 30-minute minimum cache life and explicit cache breakpoints.

Cybersecurity safeguards are beefed up across all models, with configurations tailored to each tier's capabilities. Partners participating in the limited preview have government approval, suggesting regulatory oversight shaped this rollout strategy. OpenAI plans broader availability in coming weeks once testing concludes.