OpenAI splits its GPT-5.6 launch into three tiers (Sol, Terra, and Luna) behind a government-gated rollout. What the tiering means for pricing, procurement, and model-routing strategy.
OpenAI began the broad rollout of its GPT-5.6 family on July 9, splitting the release into three models: Sol, the frontier tier built for complex coding, research, cybersecurity and computer-use work; Terra, a mid-tier model pitched at roughly GPT-5.5-level intelligence at half the cost; and Luna, the smallest and fastest of the trio. The family also introduces an "ultra" mode with a Max reasoning level and heavier use of sub-agents for long-horizon tasks.
The launch mechanics matter as much as the models. GPT-5.6 first shipped two weeks earlier as a locked-down preview to roughly 20 US-government-vetted organizations, and the public release followed only after the Commerce Department's Center for AI Standards and Innovation completed its review. Staged, government-gated frontier launches now look like the template rather than the exception, a shift with real implications for enterprise procurement timelines.
On performance, OpenAI reports a state-of-the-art 88.8% on Terminal-Bench 2.1 for Sol, rising to 91.9% in ultra mode. API pricing lands at $5/$30 per million tokens for Sol (flat versus GPT-5.5), $2.50/$15 for Terra, and $1/$6 for Luna.
Our Take
the tiering is a margin story. By holding flagship pricing flat while pushing most workloads toward Terra and Luna, OpenAI is defending share at the top of the market while competing aggressively on cost-per-task below it. For buyers, the practical question is no longer "which lab" but "which tier for which workflow", and that favours disciplined model-routing strategies over single-vendor commitments.