OpenAI's GPT-6.1 Sol Addendum Keeps Astra's Safeguards Stack

OpenAI's addendum to the GPT-6 Astra system card is dated 29 September 2026 and introduces GPT-6.1 Sol. The opening says the model delivers capabilities comparable to GPT-6 Astra, which the addendum calls OpenAI's most powerful model, and pairs that with speed and affordability. It also says GPT-6.1 Sol uses the same types of data and training as Astra. Comparison numbers for older models, a footnote says, may reflect later versions than the ones published at those models' launches.

Under the Preparedness Framework, the addendum treats GPT-6.1 Sol as Critical in cybersecurity and High in the biological and chemical domain, and below the High threshold in AI self-improvement. Because of that assessment, it says the model uses the same safeguards stack as GPT-6 Astra. On biology and chemistry, the text says the reported results did not cross the indicative Critical thresholds. On cybersecurity, it says the model does reach the Critical threshold, using the framework's definition of that level, and that the determination follows automated cyber evaluations.

A safety score that is not a capability score

On Production Benchmarks, described as challenging conversations drawn from production data, the addendum says GPT-6.1 Sol scores higher than GPT-6 Sol in five of eight categories. A footnote says the evaluations ran in a research environment or through the API, which can differ from production ChatGPT because of system prompts, tools, and effort settings. The addendum points readers to the Astra card for the full description of the safeguards stack rather than restating it.

Source: GPT-6.1 Sol system card addendum.