NVIDIA and RosettaCommons released RFD4-Proteina code and ungated weights under BSD-3-Clause. The roughly 435M-parameter model co-designs protein sequence and all-atom structure, while reward-guided inference-time search branches from higher-scoring partial candidates. Public artifacts are inspectable, but quality and 1.37–1.73x BioIR speed comparisons remain publisher-reported.
NVIDIA releases RFD4-Proteina code and weights
NVIDIA and RosettaCommons released RFD4-Proteina code and ungated weights under BSD-3-Clause. The roughly 435M-parameter model co-designs protein sequence and all-atom structure, while reward-guided inference-time search branches from higher-scoring partial candidates. Public artifacts are inspectable, but quality and 1.37–1.73x BioIR speed comparisons remain publisher-reported.
What this means for you
For computational-protein teams, the open code, weights and test data make the method directly reproducible. Biological quality still requires domain evaluation; faster sampling is not evidence that a design binds or works in vivo.
Public artifacts and validated results are different.
The recorded fact is the public work and its artifacts. Author-reported outcomes do not establish independent reproduction. Next checks include usable code, task conditions and third-party validation.
Evidence & related material
Technical details & scores
| Parameters | about 435M |
|---|---|
| License | BSD-3-Clause code and ungated weights |
| Method | all-atom sequence-structure co-design plus reward-guided inference-time search |
| Publisher Results | about 400 to 60 sampling steps versus Proteina-Complexa; 1.37–1.73x BioIR speedup on six Hopper configurations |
| Time Basis | NVIDIA RSS timestamp 2026-10-21T12:30:00Z |