Courtesy; OpenAI
OpenAI launched its most powerful model to date on Thursday, but the Trump administration shaped how it reaches the public. GPT-5.6 Sol, the flagship of the new GPT-5.6 series, is beginning a limited preview with a small group of approved partners before a broader rollout. The decision follows a direct request from the US government, which asked OpenAI to stagger access rather than release the model to everyone at once.
CEO Sam Altman disclosed the arrangement to employees on Wednesday. The request came nearly two weeks after rival Anthropic restricted its most advanced model from the market under similar regulatory pressure. OpenAI said it does not want this kind of government review process to become standard practice. The company said keeping advanced tools from users and security professionals is a cost it wants to avoid. For now, it is accepting the arrangement as a short-term measure while working with the administration on a broader framework for future releases.
What GPT-5.6 Sol can do
The GPT-5.6 lineup includes three models. Sol is the flagship. Terra offers competitive performance for everyday work at roughly half the cost of the previous generation. Luna is the fastest and most affordable option in the family.
Introducing a limited preview of GPT-5.6 Sol, our next generation frontier model, as well as GPT-5.6 Terra, a balanced model for efficient, everyday work, and GPT-5.6 Luna, a fast and affordable model for high-volume work.https://t.co/OoM83SyISN
— OpenAI (@OpenAI) June 26, 2026
Sol introduces two new capabilities. The first is a max reasoning effort setting that gives the model more time to work through difficult problems. The second is an ultra mode that uses multiple subagents to speed up complex tasks. On coding benchmarks, Sol sets a new standard on Terminal-Bench 2.1, which tests command-line planning and tool coordination. On GeneBench v1, which measures biology analysis, Sol outperforms GPT-5.5 while using fewer tokens. In cybersecurity, Sol delivers the strongest results of any OpenAI model to date. On the ExploitBench evaluation, it matched the performance of Anthropic’s Mythos Preview while using roughly a third of the output tokens.
The cybersecurity issue driving OpenAI’s phased launch
Sol’s cybersecurity capability is part of what drew government attention. OpenAI said the model is better at finding and fixing vulnerabilities than at executing complete attacks. In testing on Chromium and Firefox browsers, Sol identified bugs and exploitation building blocks but did not produce a full functional exploit on its own. OpenAI said it did not cross what the company defines as its Cyber Critical threshold.
To manage risk, OpenAI built a layered safeguard system into the launch. This includes training the model to refuse prohibited cyber assistance and running real-time output monitoring during generation. Account-level review also flags broader patterns of misuse across conversations. OpenAI dedicated more than 700,000 A100-equivalent GPU hours to automated red-teaming. The goal was to find universal attack patterns that could bypass safeguards across many different scenarios.
OpenAI pricing and what comes next
GPT-5.6 is available through the API and Codex during the preview period. Pricing is set per million tokens. Sol costs $5 for input and $30 for output. Terra runs $2.50 for input and $15 for output. Luna is $1 for input and $6 for output.
OpenAI also announced a deal with Cerebras to run GPT-5.6 Sol at speeds of up to 750 tokens per second starting in July. Access will begin with a limited group of customers as capacity grows. Broader access through ChatGPT and the API is coming in the weeks ahead. Whether the government’s involvement in future launches becomes a pattern depends on how the administration’s AI framework takes shape.
SOURCE: Bloomberg
About the Author
