發生什麼事
For now, the performance figures come from a research draft and NVIDIA's blog. The draft looks at four kinds of tasks: DNA binding, enzyme scaffolds, small-molecule binding and protein binding. NVIDIA says the model beat the comparison models in those tests on both how often its designs succeeded and how fast it produced them (NVIDIA press release). The blog adds that on a benchmark using 200-amino-acid proteins, it generates designs about 13 times as fast as RFdiffusion3 (open models blog).
There are plenty of ways to get it: Hugging Face, OpenAI, Amazon Bio Discovery and Anthropic's Claude Science all offer it already. The same press release also says Siemens Healthineers and NVIDIA are jointly developing adaptive radiation therapy and is evaluating whether Nemotron can help oncologists plan treatment (NVIDIA press release).
01Sources
- Healthcare and Life Sciences Leaders Build on NVIDIA to Advance Drug Discovery and Expand Patient Care: NVIDIA press release, published 2026-10-21 18:13 (Taipei time); RFD4-Proteina and the UW Institute for Protein Design, higher success rates and speed than the models compared across four benchmark sets, available today through partners including Amazon Bio Discovery / Claude Science / Hugging Face / OpenAI, Siemens Healthineers on adaptive radiotherapy and evaluating Nemotron
- NVIDIA Open Models Advance Europe’s Sovereign Intelligence Across Industries: NVIDIA's official blog, published 2026-10-21 18:00 (Taipei time); RFD4-Proteina roughly 13x faster than RFdiffusion3 on a 200-amino-acid protein benchmark