OpenAI publicly launched all three GPT-5.6 variants today, July 9, 2026, ending a two-week restricted preview that was limited to roughly 20 trusted partner organizations. The rollout follows clearance from the Trump administration, which required additional federal testing before approving wider availability.

The three models — Sol, Terra, and Luna — are now available to all API customers and ChatGPT subscribers at the pricing OpenAI previewed in late June.

What Changed Since the Preview

The public launch is not just a wider rollout of the same thing. Several capabilities have been confirmed that were not part of the original preview announcement.

Ultra mode is the most significant addition. Available only on Sol, Ultra mode shifts the model from a single sequential reasoning chain to a full multi-agent system. Sol decomposes complex tasks and spawns parallel subagents that work simultaneously, then synthesizes the results. On Terminal-Bench 2.1, Sol in standard mode scored 88.8 percent. Sol Ultra pushed that to 91.9 percent. For businesses running complex analytical or coding workflows, that gap matters.

Max reasoning effort is now available across all Sol workloads, giving the model extended time on difficult problems before returning an answer. Think of it as Sol’s equivalent of taking an hour to work through a hard problem rather than answering in thirty seconds.

Sol Fast mode on Cerebras is launching in parallel for select enterprise customers, reaching up to 750 tokens per second. That speed tier is priced higher ($12.50 per million input tokens, $75 per million output) but changes what is possible for real-time applications — live customer calls, rapid document review, and fast-turnaround analysis that previously required human specialists.

The Government Review Angle

The delay between preview and public launch was not a technical issue. The Trump administration’s Department of Commerce, through its Center for AI Standards and Innovation, conducted independent testing of the GPT-5.6 models before permitting broad public access. OpenAI’s technical team traveled to Washington to support the review.

This is a new operational reality for enterprise AI procurement. Government pre-clearance for frontier models introduces a validation layer that did not exist two years ago. For organizations that need to justify AI tool choices to compliance teams or regulators, a model that cleared federal review is a different kind of credential than one that simply passed the developer’s internal benchmarks.

Pricing Refresher

For teams that missed the preview announcement, here is the full pricing structure:

Model	Input (per million tokens)	Output (per million tokens)
Sol (standard)	$5.00	$30.00
Sol (Ultra / Fast)	$12.50	$75.00
Terra	$2.50	$15.00
Luna	$1.00	$6.00

Terra is the number that most enterprise teams should study. It delivers performance comparable to GPT-5.5 at half the price. For teams running large-volume AI workflows — document processing, customer support automation, data extraction — that cost difference compounds fast.

Luna at $1 per million input tokens is OpenAI’s most aggressive pricing for high-frequency, lower-complexity tasks. Call routing, inbox triage, data classification, quick summarisation: these are Luna workloads.

What This Means for Business

The cost curve just moved again. AI model pricing has been falling throughout 2026, but Terra’s pricing effectively makes GPT-5.5-class capability available at a 50 percent discount. If your team has been holding off on scaling AI-assisted operations because of per-token cost, the constraint just got smaller.

Ultra mode is the story for serious AI workflows. Businesses running multi-step agent workflows — research pipelines, code review automation, complex data analysis — have been working around the limitation that most frontier models execute steps sequentially. Ultra mode’s parallel subagent architecture changes that. It is early, but the benchmark jump suggests this is real.

Government review sets a precedent. OpenAI voluntarily shared its models with federal reviewers before public launch. Anthropic has been doing something similar with its responsible scaling policy. The pattern suggests that frontier AI labs are aligning around pre-deployment government coordination, at least for the most capable systems. Enterprise buyers should expect this to become a standard part of the procurement landscape.

Model selection just got cleaner. Sol for high-stakes agentic work. Terra for general-purpose AI operations at scale. Luna for volume processing where cost and speed matter more than peak capability. That three-tier structure is easier to apply than the previous generation’s naming conventions.

For data teams and AI practitioners already running Claude or other frontier models in production, the practical question now is whether Sol Ultra’s multi-agent architecture performs better than your current orchestration layer for complex tasks, and whether Terra’s pricing unlocks workflows that were previously too expensive to automate.

If you want to understand how to get the most out of frontier AI models in a business context — including agent orchestration, cost management, and workflow design — Enterprise DNA’s Omni Apps team builds and runs these systems for businesses that want results without the trial-and-error.