• Sources: primary, discussion
  • Summary: The addendum, published 2026-10-30, says GPT-Rosalind-6 uses the same weights as GPT-6 Astra with no added training data or post-training, and is offered only through trusted access to vetted researchers with fewer refusals on dual-use biology. At the highest reasoning budget it scored 49.6% on MedChemBench v7, 37.7% on GeneBench Pro v13 and 61.3% on LifeSciBench Gold, against 28.9%, 16.7% and 44.0% for GPT-Rosalind-5.5. OpenAI treats it as High in biological and chemical capability and Critical in cybersecurity.
  • Why it matters: Drug discovery teams get a variant defined by access policy and refusal behavior rather than new training, so GPT-6 Astra evaluations largely carry over.
  • Comments: Hacker News readers debate the trusted-access model and how vendor-run life sciences benchmarks compare with public ones.