OpenAI GPT-5.6 Limited Preview: Sol, Terra, and Luna — Pricing, Access Rules, and Government Coordination

Sol carries the same rate card as GPT-5.5 ($5/$30 per million tokens), making it an effective free capability upgrade. Terra delivers GPT-5.5-class capability at roughly half the price. Luna is the most affordable model in the lineup, designed for scale.

Eligibility and Access Rules

• Limited to approximately 20 trusted partner organizations during the preview.
• The U.S. government also has access through the API and Codex.
• Access is approved on a customer-by-customer basis, coordinated through the U.S. government.
• All three models — Sol, Terra, and Luna — are subject to the same access restrictions. The limitation applies to the entire model family, not just the flagship Sol.
• OpenAI stated: "We believe in broad access, and we plan to make GPT-5.6 available broadly soon."

Government Coordination

This rollout marks a first for commercial AI: it is the first White House-gated frontier AI commercial release.

• The rollout was coordinated by the Office of the National Cyber Director (ONCD) and the Office of Science and Technology Policy (OSTP) under a June 2 executive order.
• OpenAI briefed the U.S. government ahead of the preview and agreed to a limited start at their request.
• Access during the preview is reportedly approved by the U.S. government on a per-customer basis.
• The Trump administration requested the limitation, and OpenAI complied.

Product Exclusions

The GPT-5.6 models are not available in ChatGPT during the initial preview. Broader access across ChatGPT, Codex, and the API is planned for general availability.

Prompt Caching Changes

GPT-5.6 introduces more predictable prompt caching — a significant change for agentic loops that repeatedly send large, stable contexts (codebases, long system prompts, tool schemas) on every step.

• Caching is fully automatic with no code changes required.
• Cached input tokens are billed at 10% of the normal input price (a 90% discount).
• The default prompt_cache_retention for organizations without Zero Data Retention (ZDR) is now 24 hours instead of in_memory, enabling extended caching by default.
• Prompt caching can reduce latency by up to 80% and input token costs by up to 90%.
• GPT-5.6 adds explicit cache breakpoints and a 30-minute minimum cache life.
• Cache writes are billed at 1.25× the uncached input rate.

Troubleshooting Steps

If your organization has been granted access but you cannot access GPT-5.6, OpenAI's Help Center recommends the following steps:

1. Confirm whether your access covers the API, Codex, or both.
2. Confirm you are using the approved API organization or Codex workspace and the approved account email.
3. Try again and record the exact error message.
4. Contact your OpenAI account team with the error details, including your API organization ID or Codex workspace ID and the affected account email.

For broader API troubleshooting on the gpt-5 series, OpenAI recommends:

• Tighten API parameters: set reasoning.effort to "minimal" or "low" for routine work.
• Give the assistant an explicit stop condition and a single, fast self-check before it replies.
• Consider using a smaller model variant (e.g., gpt-5-mini) for simpler tasks.