OpenAI launches GPT-5.6 series (Sol, Terra, Luna) under government preview; Sol at $5/$30 per million tokens
OpenAI announced a limited preview of the GPT-5.6 family on June 26: Sol (flagship), Terra (balanced, 2x cheaper than GPT-5.5), and Luna (fast, lowest cost). Sol is priced at $5 input / $30 output per million tokens—roughly half Anthropic's Claude Fable 5 pricing ($10 input / $50 output). The rollout is restricted: only ~20 government-approved partner companies have access during preview, with broader availability expected in coming weeks.
Sol introduces two new reasoning modes: max (deeper inference time) and ultra (subagent-based coordination for multi-step workflows). Benchmarks highlight improvements in coding (Terminal-Bench 2.1 state-of-the-art), biology, and cybersecurity, where Sol matches Anthropic's Mythos Preview while using ~1/3 the output tokens. OpenAI emphasizes safety: 700,000+ A100-equivalent GPU hours spent on red-teaming. Sol can identify and patch vulnerabilities but does not autonomously execute end-to-end attacks under evaluated conditions.
The preview-before-general-availability pattern signals new norms for frontier releases. The Trump administration's Office of the National Cyber Director and Office of Science and Technology Policy requested OpenAI pre-launch the models with government partners. For practitioners, the cost ladder (Luna ≈$1/$6, Terra ≈$2.50/$15, Sol ≈$5/$30) is designed for routing: use Luna for volume, Terra for everyday tasks, Sol for hard work. Cache improvements (1.25x write, 90% read discount, 30-min minimum TTL) benefit agentic loops with reused long context.