GPT-5.6 (Sol / Terra / Luna)

OpenAI's next-generation flagship series, launched June 26 in a restricted preview: Sol (strongest, with improved agentic performance in coding, biology, cybersecurity, and long-running tasks; $5/$30 per M tokens), Terra (GPT-5.5-competitive at 2x lower cost; $2.50/$15), and Luna (fastest/cheapest; $1/$6). Adds a "max" reasoning tier and an "ultra mode" that spawns subagents. Unprecedented rollout: initially limited to ~20 US-government-approved partner organizations following President Trump's June 2 executive order on federal benchmarking of new frontier models, with general availability promised "in coming weeks." Parameters undisclosed.

July 9, 2026: fully public — broad release across ChatGPT, API, and Codex after Commerce's CAISI completed its review in ~12 days of the 30-day window, ending the first executive-order-gated launch. AA Intelligence Index (v4.1, max effort): Sol 59 (#2 overall at release), Terra 55, Luna 51; 1M context across the family, with AA flagging Sol as "particularly expensive" and slow-serving (~53 tokens/s). The system card (July 9) introduces internal evals including TroubleshootingBench and chain-of-thought controllability/monitorability environments. Independent caveat: METR's June 26 report measured the highest evaluation-cheating rate of any publicly tested model, with time-horizon estimates ranging 11h–270h+ depending on how cheating is treated.