OpenAI Previews GPT-5.6 Sol Under Government Access Limits

OpenAI previewed its GPT-5.6 model family on June 26, 2026, and called the flagship its strongest model yet. The catch is that almost no one can use it.

According to OpenAI, the release went first to a small group of trusted partners through the API and Codex, with a broad launch planned in the coming weeks. The company said it took this step at the request of the US government.

OpenAI Split GPT-5.6 Into Three Named Tiers

GPT-5.6 arrives as three models rather than one. OpenAI named them Sol, the flagship, Terra, a balanced everyday model, and Luna, the fastest and lowest-cost option.

The naming is a deliberate break from version numbers. As VentureBeat reported, the number now marks the generation while Sol, Terra and Luna mark durable capability tiers that can advance on their own cadence.

Terra performs competitively with GPT-5.5 at roughly half the cost, and Luna delivers strong capability at OpenAI's lowest price point. The split is meant to turn model choice into a routing decision, matching each job to the right tier.

Sol Set New Benchmark Highs in Coding and Cyber

Sol posted record numbers on the evaluations OpenAI chose to share. Cyberpress reported that Sol set a state-of-the-art result on Terminal-Bench 2.1, scoring 88.8 percent on command-line workflows, with an ultra mode pushing that figure to 91.9 percent.

The model also gained ground in security work. On the ExploitBench benchmark, Engadget noted that Sol came close to Anthropic's Mythos Preview while using only about one-third of the output tokens.

OpenAI also introduced two new controls. A max reasoning effort gives Sol more time to think on hard problems, and an ultra mode uses subagents to accelerate complex, long-horizon work. These tools matter most for the agentic coding and research tasks the broader AI automation and agents market keeps pushing toward.

The Government Gated the Launch

The most consequential detail is who can access the model. Axios reported that GPT-5.6 went out as a limited preview to around 20 companies whose participation was approved by the government, with wider access expected to follow.

The arrangement traces to a June 2, 2026 executive order that asks AI companies to present their most powerful models for voluntary government review before public release. Engadget reported that under the order, labs submit covered models roughly 30 days ahead of availability.

OpenAI did not hide its discomfort. In its announcement the company wrote that it does not believe this kind of government access process should become the long-term default, arguing it keeps the best tools from developers, enterprises and cyber defenders who need them.

Pricing Holds Steady at the Top

OpenAI kept Sol's price in line with the prior flagship. Cyberpress listed Sol at 5 dollars per million input tokens and 30 dollars per million output tokens, with Terra at 2.50 dollars input and 15 dollars output, and Luna at 1 dollar input and 6 dollars output.

Cache reads receive a 90 percent discount, while cache writes are billed at 1.25 times the uncached input rate. OpenAI also said it is working with Cerebras to target up to 750 tokens per second for Sol beginning in July 2026.

The staggered launch arrives during the busiest stretch of model releases this year, a pace that keeps reshaping AI compute economics across the industry. The competitive gap at the model layer now narrows by the week.

What Enterprises Should Watch

The classification matters as much as the benchmarks. VentureBeat reported that OpenAI is rating all three GPT-5.6 models, not just Sol, at its High risk level for both cybersecurity and biological and chemical capability.

That dual-use framing is why the access gate exists, and why buyers should treat availability dates as provisional. The same tension over who controls advanced AI is playing out in the wider debate over who gets to police AI.

For now the practical move is patience plus preparation. Teams that map workloads to the Sol, Terra and Luna tiers ahead of general availability will be ready to migrate the moment the gate opens, rather than scrambling when broader access finally lands.