OpenAI confirmed that GPT-5.6 Sol, Terra, and Luna will launch publicly this Thursday, ending a two-week government-gated preview and opening its most capable model family to developers worldwide. Here is who built it, what changed, when it lands, where it costs what, and why the rollout was restricted in the first place.
- What Is Launching This Thursday
OpenAI’s official post on X states plainly: GPT-5.6 Sol, along with Terra and Luna, will launch publicly this Thursday, and preview access is expanding globally starting now. That single line answers the question most developers have been asking since late June: when does the gate actually open.
The announcement follows weeks of restricted access. OpenAI had released three versions of GPT-5.6 under the names Sol, Terra, and Luna, with Sol positioned as the most powerful, Terra balancing efficiency and power, and Luna built for speed and affordability. Until this week, that lineup sat behind a government-coordinated approval process rather than a standard staged rollout.
- Sol, Terra, and Luna: What Each Model Actually Does
The three names are not marketing flourishes. In OpenAI’s new naming system, the number identifies the model’s generation while the name identifies a durable capability tier, so each tier can advance on its own schedule going forward. That is a structural shift from the old GPT-4/GPT-5 numbering, and it means future updates to, say, Luna won’t force a rename of the whole family.
- Sol: The Flagship Reasoning Model
Sol is the strongest and most capable tier, the one OpenAI leads with on every benchmark in this release, and the only model that unlocks the new max reasoning effort setting and ultra mode. It targets the hardest multi-step problems, long-horizon coding, and cybersecurity research, at the highest per-token cost in the family.
- Terra: The Everyday Production Model
Terra delivers competitive performance against GPT-5.5 while costing about half as much, making it the practical default for high-volume business tasks such as document analysis, internal tools, and customer support workflows.
- Luna: Speed and Cost Efficiency
Luna brings strong capability at the lowest cost point in the family, aimed at routine automation, summarization, and drafting tasks where raw reasoning depth matters less than speed and price.
GPT-5.6 also introduces more predictable prompt caching with explicit cache breakpoints and a 30-minute minimum cache life; cache writes are billed at 1.25 times the uncached input rate, while cache reads keep the standard 90 percent discount. For teams running repetitive agentic prompts, that caching structure can meaningfully cut real-world spend below the headline per-token prices.
- Why the Rollout Was Restricted in the First Place
This is the part of the GPT-5.6 story that made headlines beyond the usual AI press. Prior to the June 26 launch, reporting indicated OpenAI agreed to stagger the release after the Trump administration formally requested controlled access, citing national security concerns, a request that followed a high-profile incident in which the administration’s export control directives forced Anthropic to take its Fable 5 and Mythos 5 models entirely offline.
OpenAI itself framed the decision as being made at the government’s request, starting with a limited preview for a small group of trusted partners whose participation had been shared with the government, before releasing the models more broadly. The company was candid that it does not see this kind of government-mediated access process as an ideal or sustainable long-term default, arguing it keeps useful tools out of the hands of users, developers, and cyber defenders who need them.
OpenAI released GPT-5.5-Cyber shortly before the GPT-5.6 series as part of its Daybreak initiative, a model built for automated vulnerability detection and patch generation, which set the stage for GPT-5.6 Sol’s much stronger offensive-defensive cyber capability profile. Regulators wanted a closer look before that capability reached the open market.
- GPT-5.6 vs. the Competition
Early access reports from the preview period describe Sol writing tighter, more efficient code than Claude Opus 4.8, at roughly half the per-token cost of Claude Fable 5, while leading the published agentic coding benchmark — though, as above, those benchmark claims carry the METR reward-hacking caveat. Claude Fable 5 itself returned to global availability on July 1 after a 19-day forced suspension under U.S. export controls, and moved from subscription-included access to paid usage credits at $10 per million input tokens and $50 per million output tokens on July 7 — meaning Sol currently undercuts Fable 5 on price even as both compete at the frontier tier.
- Timeline: How We Got Here
June 26, 2026: OpenAI previews GPT-5.6 Sol, Terra, and Luna, restricted to roughly 20 government-vetted partner organizations via API and Codex only.
Late June–early July: Independent reporting draws a parallel between GPT-5.6’s gated rollout and the export-control situation that took Anthropic’s Fable 5 and Mythos models offline earlier in June.
July 6, 2026: An OpenAI Codex engineer confirms Sol Ultra will ship inside the Codex client.
July 7–8, 2026: Prediction markets converge on a public launch within days; OpenAI confirms via its official account that GPT-5.6 Sol, Terra, and Luna launch publicly this Thursday.
What Happens Next
The only confirmed timeline from OpenAI has been that the preview would expand after further testing, with more specific commitments remaining unofficial until this week’s confirmation. With Thursday’s public launch now confirmed, the next milestones to watch are ChatGPT integration timing, Cerebras-hosted Sol capacity for enterprise customers, and whether OpenAI publishes the fuller, independently-verified evaluation suite it has promised.
Frequently Asked Questions
1. When does GPT-5.6 Sol launch publicly?
OpenAI confirmed that GPT-5.6 Sol, Terra, and Luna launch publicly this Thursday, expanding preview access globally after roughly two weeks restricted to about 20 government-vetted partner organizations.
2. What is the difference between GPT-5.6 Sol, Terra, and Luna?
Sol is OpenAI’s flagship model built for frontier reasoning, coding, and cybersecurity work. Terra is a balanced mid-tier model priced about half of GPT-5.5 with comparable performance. Luna is the fastest and cheapest tier, designed for high-volume, everyday tasks.
3. How much does GPT-5.6 cost?
Per one million tokens, Sol costs $5 for input and $30 for output, Terra costs $2.50 for input and $15 for output, and Luna costs $1 for input and $6 for output. Cache writes are billed at 1.25 times the standard input rate, while cached reads keep a 90 percent discount.
4. Why was GPT-5.6 restricted before this launch?
OpenAI staged the rollout after the U.S. government requested a controlled preview, citing cybersecurity and national security concerns tied to the model’s strengthened coding and vulnerability-research capabilities.
5. Is GPT-5.6 available in ChatGPT?
During the preview period, GPT-5.6 was limited to the OpenAI API and Codex. Thursday’s public launch is expected to widen that access, though OpenAI has historically staggered ChatGPT integration separately from API availability.
6. What is Sol Ultra mode?
Ultra mode is a multi-agent configuration where Sol spawns parallel subagent processes to tackle different parts of a complex task at once, rather than reasoning through one sequential chain. It pushed Sol’s Terminal-Bench 2.1 score from 88.8 percent to 91.9 percent, though it also multiplies token cost.