OpenAI Launches GPT-5.6 to the Public, Ending Its Government-Gated Preview
OpenAI moved its GPT-5.6 model family to general availability on July 9, after federal reviewers cleared the release that had restricted Sol, Terra, and Luna to trusted partners.
Editor's Note ·
- Correction:
- The article quotes ChatGPT Work as "constructed on Codex and powered by GPT-5.6." This phrase does not appear verbatim in the cited Thurrott article. Thurrott's own commentary (not an OpenAI quote) instead says, in two separate sentences, that ChatGPT Work "is built on Codex" and separately "It's powered by the GPT-5.6 frontier model." The quoted phrase is a paraphrase and should not have been presented in quotation marks.
- Correction:
- The article states the June executive order "requir[es] companies to submit powerful models for government review 30 days before public release." The cited Engadget source describes the order as one that "asks companies to voluntarily present their most powerful models for government review" 30 days before release — a voluntary framework, not a mandatory requirement. The Machine Herald's own July 5 article on the same order correctly described it as "requesting that AI companies voluntarily submit" their models.
Overview
OpenAI has made its GPT-5.6 model family generally available, moving the three-tier lineup — Sol, Terra, and Luna — out of a government-restricted preview and into open release on July 9, 2026, according to MarkTechPost. The models had been available only to “a small group of trusted partners” since a limited preview began in late June, before the Trump administration granted permission for wider release following additional government testing, according to Engadget. The Machine Herald previously reported on the restricted preview when it began.
What We Know
GPT-5.6 spans three pricing tiers. Sol, the flagship, costs $5 per million input tokens and $30 per million output tokens; Terra costs $2.50 and $15; and Luna, the cheapest tier, costs $1 and $6, according to MarkTechPost. Terra is positioned to deliver performance similar to GPT-5.5 at roughly half that model’s cost, according to Engadget.
Federal reviewers at the Commerce Department’s Center for AI Standards and Innovation evaluated the model family before the government cleared it for broader release, according to Engadget and SiliconANGLE. The review stems from a June executive order from President Trump requiring companies to submit powerful models for government review 30 days before public release, according to Engadget; OpenAI said it sent technical experts to Washington to address the government’s questions, per the same report. OpenAI has been open about its discomfort with the arrangement. “We don’t believe this kind of government access process should become the long-term default,” the company said, according to Engadget.
OpenAI paired the GPT-5.6 rollout with two companion launches. ChatGPT Work, an agent “constructed on Codex and powered by GPT-5.6” that gathers information across a user’s apps and workflows “to create finished materials like sheets, slides, docs, and web apps,” began rolling out to Pro, Enterprise, and Edu subscribers immediately, with Plus and Business tiers to follow within days, according to Thurrott. Separately, OpenAI introduced GPT-Live-1 and a smaller GPT-Live-1-mini, voice models built on a full-duplex architecture, with GPT-Live-1 set as the default voice option in paid ChatGPT plans and GPT-Live-1-mini powering the free tier, according to SiliconANGLE.
OpenAI is also gating some of GPT-5.6’s more sensitive capabilities behind stronger identity checks. Individual account holders must enable Advanced Account Security with hardware-backed passkeys by September 1 to retain access to the models, which OpenAI restricts through its Trusted Access for Cyber program, according to iClarified. On OpenAI’s own cybersecurity benchmark, ExploitBench2, Sol scored 73.5 percent, per the same report.
On other benchmarks, OpenAI says Sol scored 53.6 percent on the Agents’ Last Exam, a test of long-running professional workflows, beating Anthropic’s Claude Fable 5 by 13.1 percentage points, according to iClarified. But OpenAI’s own released figures also show a gap in the other direction: on SWE-Bench Pro, Sol scored 64.6 percent against 80.3 percent for Anthropic’s Claude Mythos 5 and 80 percent for Claude Fable 5, a roughly 15-point deficit, according to MarkTechPost.
What We Don’t Know
Neither Engadget nor SiliconANGLE detailed what the Commerce Department’s evaluation of GPT-5.6 actually found, or the specific criteria that determined the model was ready for wider release. It also remains unclear whether this kind of pre-release federal review will apply to OpenAI’s future model launches or was specific to GPT-5.6.
Analysis
The GPT-5.6 rollout follows a pattern that has now played out with two frontier AI labs in as many months. Anthropic’s Mythos 5 was similarly re-released to trusted partners after a government export block, as The Machine Herald previously reported. With OpenAI now completing a comparable review-and-release cycle for GPT-5.6, the open question — one OpenAI itself has raised — is whether pre-release federal review of frontier models becomes a recurring checkpoint on future launches or remains a one-off response to this generation of releases.