OpenAI's GPT-5.6 Sol, Terra, and Luna Clear Government Review, Set for Public Release Thursday

Since OpenAI announced GPT-5.6 on June 26, the model family has been available only to a "small group of trusted partners" whose participation, per OpenAI's own blog post, was shared with the U.S. government. That restricted phase ends Thursday.

CEO Sam Altman confirmed the timeline in a post on X: "GPT-5.6 sol launches thursday! happy building."

What These Models Actually Do

GPT-5.6 comes in three tiers. Sol is the flagship, built for deep reasoning, coding, biology, and cybersecurity tasks. OpenAI says Sol makes fewer factual errors than its predecessors and introduces a new "ultra mode" that deploys subagents to accelerate complex work, according to the company's blog post.

Terra is the middle-ground option: roughly competitive with GPT-5.5 in capability but priced at half the cost. Luna is the cheapest and fastest to run. Pricing, per Engadget: Sol runs $5 per million input tokens and $30 per million output; Terra costs $2.50 input and $15 output; Luna costs $1 input and $6 output.

Benchmark data from Wikipedia's GPT-5.6 entry shows Sol scoring 88.8% on TerminalBench 2.1, edging out rival Anthropic's Claude Mythos 5 at 88.0%. Sol Ultra reaches 91.9%. Luna scores 82.5%, which trails GPT-5.5's 83.4% on the same benchmark.

How the Government Review Actually Worked

President Trump signed an AI executive order in early June directing the Defense Department to design a voluntary system under which AI developers share frontier models with the government 30 days before public release. Federal agencies were given 60 days to develop an evaluation process, according to CNBC.

The entity that reviewed GPT-5.6 was the Center for AI Standards and Innovation (CAISI), housed inside the Department of Commerce. According to ZeroHedge, citing Axios, OpenAI sent technical experts to Washington during the review period to answer questions in real time. The Trump administration gave OpenAI the green light for a wider release after additional testing and meetings were completed.

One important nuance: a White House official told CNET that the government did NOT formally give OpenAI "the green light" in a legal sense, and that such approval was not required under the executive order. All engagement remains voluntary. CNET is the only outlet that included this clarification. The framing from ZeroHedge and Engadget implies a formal government approval gate that, technically, does not yet exist as binding law.

OpenAI itself said in June that it does not want this review process to become "the long-term default," arguing it "keeps the best tools from users, developers, enterprises, cyber defenders, and global partners who need them."

The Tension Worth Taking Seriously

Critics of the review process have a legitimate concern. The Trump administration spent years arguing that AI regulation would slow U.S. innovation and hand China a strategic advantage. Now the same administration is asking AI companies to hold their most powerful models for government assessment before public release. This represents a shift in approach. CNET's framing on this point is accurate: it changes the pace of model releases and affects the U.S.'s competitive position on the global AI stage.

The counter-argument, which OpenAI has effectively embraced, is that voluntary early access builds trust and helps establish evaluation standards before a binding framework is imposed. OpenAI said in June it's working with the government to create "a repeatable process for future model releases." Getting ahead of a harder mandate by cooperating on a softer one is a reasonable calculation for a company that needs federal goodwill.

Neither position is obviously wrong. The review added roughly two weeks to the public rollout: GPT-5.6 was announced June 26 and goes public July 9.

Anthropic as Context

OpenAI's situation is notably smoother than what Anthropic went through. Anthropic was directed to block all access to its Claude Fable 5 and Mythos 5 models under a separate legal authority: a Department of Commerce export control directive aimed at preventing foreign nationals from accessing the models. That is a harder legal instrument than the voluntary AI executive order. The Commerce Department eventually lifted the directive, and Anthropic has since restored access to both models, per CNBC and Engadget.

The legal distinction matters. OpenAI was never operating under a mandatory block. Anthropic was. Conflating the two situations, as some coverage has done, overstates the legal pressure OpenAI faced.

GPT-Live Launches Today

Separate from the GPT-5.6 rollout, OpenAI announced GPT-Live on Wednesday. Two versions, GPT-Live-1 and GPT-Live-1 mini, are rolling out to ChatGPT users globally as of today, July 8. The models listen and speak simultaneously, which OpenAI says makes interaction feel "much more like having a real conversation," per the company's blog post.

What Happens Next

The open question sitting behind all of this is the 60-day clock. Trump's June AI executive order gave federal agencies 60 days to develop a formal evaluation process. Whatever CAISI produces by then will determine whether today's voluntary, relationship-based review becomes something with harder edges and how fast the next generation of frontier models reaches the public.