In detail
- Sol outperforms Anthropic's Claude Mythos 5 on agentic coding (88.8% vs. 88% on Terminal-Bench 2.1) and costs half as much as Fable 5 ($5/$30 per million tokens vs. $10/$50).
- Government approves access "customer by customer" during preview; OpenAI hopes for broad release in "a couple of weeks" if all goes smoothly.
- Sol features "most robust safety stack" with 700,000 A100e GPU hours of automated red-teaming; model refuses to carry out end-to-end cyber attacks.
- OpenAI states publicly: "We don't believe this kind of government access process should become the long-term default"—argues it keeps best tools from users, developers, and cyber defenders.
Why it matters
The forced delay reveals how the US government has effectively established a de facto licensing regime for frontier AI without clearly defined safety standards. For German SMEs, this means access to leading US AI models is becoming less predictable, and pressure to adopt European or open-source alternatives is mounting.
For you Monitor whether this approval process becomes standard practice—it may force you to shift toward European or open models or plan for longer deployment timelines.
Updates
OpenAI previews GPT-5.6 Sol with stronger capabilities in coding, science, and cybersecurity, but must phase its release after the Trump administration demands customer-by-customer government approval.
- Model shows improved performance in coding, science, and cybersecurity with an advanced safety stack.
- US government (Office of the National Cyber Director, Office of Science and Technology Policy) requires limited preview phase with individual approval per customer.
- Commerce Secretary Howard Lutnick warned OpenAI against release without sign-off from multiple agencies.
- Altman hopes for broader release after a few weeks if preview succeeds; criticizes the approval model as unsustainable long-term.
OpenAI previews GPT-5.6 Sol with stronger capabilities in coding, science, and cybersecurity, but must initially limit access to select partners as the US government approves customers individually.
- Model demonstrates enhanced performance in coding, science, and cybersecurity paired with an advanced safety stack.
- US government (Office of the National Cyber Director, Office of Science and Technology Policy) requires limited release to small partner group during preview phase.
- Sam Altman expects broader release "a couple of weeks later" if preview succeeds; Commerce Secretary Howard Lutnick warned against release without approval from additional agencies.
- OpenAI states this is "not our preferred long term model" and commits to working with government on a more sustainable approach for future releases.
OpenAI unveils GPT-5.6 Sol with improved capabilities in coding, science, and cybersecurity, but must phase its release under US government oversight—each customer requires separate approval.
- GPT-5.6 Sol demonstrates stronger performance in coding, science, and cybersecurity with an advanced safety stack.
- US government (Office of the National Cyber Director, Office of Science and Technology Policy) mandates customer-by-customer approval during preview; Altman expects broader release in "a couple of weeks" if all goes well
- Commerce Secretary Howard Lutnick warned OpenAI against release without sign-off from multiple agencies; OpenAI states this is not its preferred long-term model.