BoltzGen Arrives at Ariax Bio: Design Antibodies and More
Two weeks ago, the MIT team behind the Boltz molecular foundation models released BoltzGen, their most ambitious model yet. While Boltz-1 and Boltz-2 proved open-source models could match AlphaFold3 for structure prediction, BoltzGen ventures into protein design, creating new proteins to bind any biomolecular target. Today, we're bringing BoltzGen to the Ariax platform as our second protein design engine alongside BindCraft.
If you're new to Ariax, learn about our mission, origin story, and transparent subscription-free business model in Introducing Ariax Bio.
Antibodies: The Final Frontier for AI-Powered Protein Design
While BindCraft revolutionized de novo miniprotein design, antibodies remain the pharmaceutical industry’s dominant therapeutic modality for good reason. They benefit from decades of clinical precedent, established manufacturing pipelines, and well-characterized development pathways, while sidestepping the immunogenicity risks posed by synthetic miniproteins.
Historically, antibodies have defied computational design. They bind targets through flexible complementarity-determining region (CDR) loops that are far harder to model than the predictable α-helices and β-sheets that comprise miniproteins. Even the leading model AlphaFold3 correctly predicts antibody-antigen complexes only ~50% of the time. As a result, in silico antibody design success rates have languished below 1% while other modalities like miniproteins and peptides have taken off.
That landscape shifted dramatically when Chai Discovery unveiled Chai-2 earlier this year. The model achieved double-digit experimental success rates for de novo antibody design, marking a major breakthrough, but also creating a rare case where proprietary tools outpaced open-source alternatives.
Open-Source Strikes Back
Fortunately, the BoltzGen team answered this challenge. Like Chai-2, BoltzGen designs antibodies (in this case, VHH single-domain antibodies) beyond miniproteins, but remains completely open-source under an MIT license, ensuring broad access for academic and commercial users.
Critically, BoltzGen achieves top-tier, state-of-the-art performance. On a set of standard benchmark targets (IL7-RA, PDL1, PDGFR, Insulin, and TNFa), BoltzGen delivered nanomolar binders for 4 out of 5 targets, with per-binder success rates approaching 60%.
More impressively, BoltzGen was also rigorously tested on genuinely novel targets. The BoltzGen team selected nine proteins with no similar sequences (>30% identity) in any bound structure across the entire Protein Data Bank, ensuring the model couldn’t rely on memorized binding modes. By testing just 15 nanobody designs per target, BoltzGen yielded nanomolar-affinity binders for six targets: a 66% success rate. Protein binders performed identically with 66% target-level success with nanomolar affinities.
Beyond antibodies, BoltzGen demonstrates remarkable versatility: functional binders for diverse bioactive peptides (melittin, indolicidin), peptides targeting intrinsically disordered protein regions in live cells, and both linear and cyclic peptides, as well as “stapled” alpha helical peptides called helicons. Its all-atom architecture also further unlocks exciting capabilities: design binders against small-molecule drugs, metabolites, RNA/DNA sequences, peptide hormones, and even unstructured protein regions.
Computational Requirements Made Accessible
Like BindCraft, BoltzGen demands substantial resources. The BoltzGen team recommends generating 10,000–60,000 designs per run in order to boil those down to a handful of synthesis candidates. This requires dozens to hundreds of GPU hours depending on target size.
Ariax’s model eliminates the barrier: pay only for compute, with no subscription or platform fees. Running BoltzGen on Ariax costs less than provisioning equivalent instances on AWS, Azure, or Google Cloud. Furthermore, with Ariax’s Turbo Mode function, you can accelerate jobs up to 8x by parallelizing across multiple GPUs, condensing days-long campaigns into an afternoon.
We are also proud to make running complex BoltzGen jobs as easy as possible. On the Ariax platform, we ensure that your job submissions are properly tailored so you can achieve your design goals. As always, your results stream to a private cloud folder where you can access and view your files at any time. Most importantly, your jobs are confidential, and the resulting sequences are 100% your IP.
BoltzGen vs. BindCraft: When to Use Each
We recommend starting with BindCraft for miniproteins and helical peptides when all-atom design is not required. The breakthrough that inspired Ariax continues to set the benchmark for miniprotein and helical peptide design, with a battle-tested architecture validated across numerous competitions and independent studies. For most users, BindCraft remains our primary recommendation for these modalities: its documented success rates are simply unmatched.
BoltzGen complements this by venturing into spaces where BindCraft wasn't architected to tread. Its sophisticated all-atom framework excels at designing VHH nanobodies, cyclic peptides, and helicons in addition to miniproteins and unmodified peptides. BoltzGen further distinguishes itself by handling target classes beyond proteins: nucleic acids, small molecules, peptide hormones, and disordered protein regions that lack stable structure. When your therapeutic vision requires antibodies or ventures beyond traditional protein-protein interactions, BoltzGen provides the versatility to execute it.
That said, BoltzGen also delivers excellent results for miniproteins and peptides, making it a valuable alternative if BindCraft struggles with a particularly challenging target. The two engines embody different architectural philosophies and it is possible to see success with one where the other fails. View them as complementary problem-solvers rather than replacements.
Available Now
BoltzGen is live on Ariax with the same streamlined interface as BindCraft: upload your target, configure parameters, and generate. Results stream to your project folder in real time; clone successful projects as templates; deploy Turbo Mode for faster iteration. Between BindCraft for miniproteins and BoltzGen for antibodies and beyond, researchers now have comprehensive, institution-grade design tools without costly licenses or IP entanglements.
Ready to design your first antibody? Launch BoltzGen on Ariax →
BoltzGen is available immediately on all Ariax accounts. For technical details and local installation, visit the BoltzGen GitHub repository. For the full manuscript and experimental data, see the BoltzGen paper.