The US Commerce Department authorised the widespread rollout of OpenAI's GPT-5.6 model following supplemental government testing
The Centre for AI Standards and Innovation conducted the evaluation with technical experts from OpenAI in Washington
The OpenAI GPT-5.6 series includes the flagship Sol, the balanced Terra, and the affordable Luna models
US commerce officials have authorised the widespread rollout of OpenAI's GPT-5.6 system following additional government testing, Reuters reported. This clearance comes under Washington's new regulatory framework for cutting-edge artificial intelligence.
According to the report, OpenAI will publicly release GPT-5.6 on Thursday following supplemental testing and meetings between OpenAI and US government officials.
1 June 2026
Get the latest issue of Outlook Business
The Center for AI Standards and Innovation conducted the evaluation. To address technical queries, the developer dispatched specialists to the capital.
Delay and Security Scrutiny
OpenAI postponed the general release of GPT-5.6 in June following a request from federal officials. The firm restricted early access to a select group of approved partners and disclosed their identities to the government.
US officials have tightened oversight of advanced model releases to detect potential risks. The move stems from fears that rival nations like China and Russia might exploit the technology for espionage or military operations. By reviewing these systems early, Washington hopes to spot hazards like cyberattacks before public deployment.
What Does GPT 5.6 Have to Offer?
According to the AI giant, the GPT-5.6 series includes Sol as the flagship model, Terra for balanced everyday tasks, and Luna as a fast, affordable option. Terra has competitive performance to GPT-5.5 while being twice as cheap and Luna brings strong capability at its lowest cost.
The company website highlights performance upgrades, safety measures, and pricing details.
Smarter, more autonomous problem-solving:
Sol is being called the company's strongest model to date. It comes with a "max reasoning effort" mode that lets the AI spend more time working through difficult problems, along with an "ultra mode" that deploys multiple sub-agents in parallel to speed up complicated tasks.
On the genomics front, Sol reportedly outperforms GPT-5.5 on the GeneBench v1 benchmark while using less computing power to get there. The company also claims Sol performs as well as Mythos on the ExploitBench cybersecurity benchmark, but using roughly a third of the output tokens Mythos needs.
Multiple layers of safety checks:
Despite the performance gains, the company says Sol stays below the "Cyber Critical" risk threshold set out in its Preparedness Framework. To keep it that way, the model relies on several layers of protection: built-in refusals at the model level and real-time filters that flag potential misuse related to cyberattacks or biological threats. On top of this, account-level monitoring helps tell apart genuine bad actors from security researchers doing legitimate dual-use work.
The company also says it put in more than 700,000 hours of A100-equivalent computing power toward automated red-teaming - essentially, running large-scale tests to hunt for jailbreaks that could work across different scenarios, so the system holds up against real attacks. This automated testing is backed up by human security experts running their own checks.
What it costs and how to access it:
Pricing is tiered by model. Sol runs at $5 per million input tokens and $30 per million output tokens. Terra is priced at $2.50 for input and $15 for output, while Luna comes in cheapest at $1 for input and $6 for output.
The update also brings in defined cache breakpoints and guarantees cached data stays usable for at least 30 minutes. Writing to the cache costs 1.25 times the standard input rate, but reading from it still comes with a 90% discount.
The Regulatory Framework
A June executive order by US President Donald Trump prompted the postponement. The directive created a voluntary system allowing creators to submit "covered frontier models" to federal officials for a maximum of 30 days prior to sharing them with approved partners.
Anthropic stated last week that the Commerce Department lifted restrictions on access to its most advanced Fable and Mythos AI models. The decision came less than three weeks after the government ordered the company to suspend the availability of the systems over national security risks.
