OpenAI is bringing its new GPT-5.6 model to market on Thursday – but only after a forced delay by the US government. The model was announced in late June but was initially restricted to select partners after the US Department of Commerce applied pressure. Only after additional testing by the Center for AI Standards and Innovation did the department approve the public launch, according to Axios.
Key Facts
- GPT-5.6 Sol Ultra achieves 91.9 percent on the TerminalBench 2.1 coding benchmark – outperforming Anthropic's Claude Mythos 5 at 88.0 percent
- Pricing: OpenAI charges $5/$30 per million tokens (input/output), Anthropic $10/$50 – OpenAI significantly cheaper
- Government approval came after additional testing by the Center for AI Standards and Innovation
- Binding standards for future model approvals still don't exist, despite Trump administration demands
Who Controls AI Releases Now?
The delay marks a turning point: for the first time, it's publicly confirmed that the US government delayed the release of a frontier model. OpenAI openly criticized the measure, arguing that the delay kept the best tools away from developers and companies. At the same time, it remains unclear by what criteria such approvals will be granted in the future – binding standards, as demanded by Trump's latest AI executive order, still don't exist.
Technical Superiority, But No Clear Rules Yet
On benchmarks, OpenAI's new generation shows clear advantages. On the TerminalBench 2.1 test, Sol Ultra dominates with 91.9 percent, while Google's Gemini 3.1 Pro Preview trails significantly at 70.7 percent. Particularly noteworthy: on cybersecurity tasks, Sol matches Claude Mythos 5 but uses only one-third of the tokens – an enormous efficiency gain.
| Model | TerminalBench 2.1 | Cost (Input/Output per Million Tokens) |
|---|---|---|
| Sol Ultra | 91.9 % | $5 / $30 |
| Claude Mythos 5 | 88.0 % | $10 / $50 |
| Gemini 3.1 Pro Preview | 70.7 % | – |
The cost savings are substantial: Anthropic's Fable 5 costs nearly double and likely becomes even more expensive through higher token consumption.
What This Means for German Companies
The news signals that government approval processes for powerful AI models are becoming the new normal – at least in the US. German and European firms should expect such delays could become standard here too, especially since the EU AI Act already defines requirements for high-risk AI. At the same time, an opportunity emerges: companies that now adopt more cost-efficient and powerful models like GPT-5.6 can gain competitive advantages in automating complex tasks – provided the regulatory landscape in Germany and the EU remains clear.
Sources
Editorially owned by Ideal Syka. Sources and method: Newsroom & method. Tips and corrections: ai@i6eal.de.
