OpenAI released supplementary notes on the GPT-Rosalind-6 system card
The story so far
OpenAI released supplementary notes for the GPT-Rosalind-6 system card, which uses the same weights as GPT-6 Astra and has not had training data added or additional post-training performed. In the GeneBench Pro v13 test, GPT-Rosalind-6 achieved a single-trial success rate of 37.7% under the highest inference budget (129 questions), surpassing GPT-Rosalind-5.5's 16.7%. The model was rated high in capability in the fields of biology and chemistry and is continuously monitored to adjust safety measures. OpenAI emphasized that the model will still refuse or respond safely to requests that may promote severe biological weaponization, and requires GPT-Rosalind-6 customers to maintain appropriate safety controls. OpenAI designed GPT-Rosalind-6 to be more suitable for legitimate advanced biological research than general available ChatGPT models. The company stated it will continue to monitor safety signals, customer experience, and emerging abuse patterns, and may update GPT-Rosalind-6's safety measures based on deployment progress.
AI-written from the reports · Updated 19 days ago
Coverage timeline
Follow the coverage to see every side of the story.
- OpenAI Deployment SafetyGPT-Rosalind-6 System Card Addendum
GPT-Rosalind-6 uses the same weights as GPT-6 Astra, no training data has been added or additional post-training performed. In the GeneBench Pro v13 test, GPT-Rosalind-6 achieved a single attempt success rate of 37.7% on the highest reasoning budget (129 questions), surpassing GPT-Rosalind-5.5's 16.7%. The model has been rated high-capability in the fields of biology and chemistry and is continuously monitored to adjust safety measures.