HF's DNA model evals: Carbon-Mid-300B nears Evo2 with just 160B training tokens
_lewtun · x · 2026-10-23
Loubna Ben Allal of Hugging Face shared evals comparing the team's new DNA models against Evo2. On pure DNA benchmarks, Carbon-Mid-300B scores 0.83 AUROC on BRCA2 VEP, 0.89 on ClinVar coding and 0.94 on ClinVar non-coding — matching or beating Carbon-3B and Evo2-7B, and far above the Qwen3-14B Instruct baseline (0.45–0.50) after midtraining.
She notes DNA performance isn't saturated yet: the new model was trained on only 160B DNA tokens versus 1T for the previous Carbon model on the same data, hinting the next version will train much longer.
More from Research
- ThinkingCap squeezes 38.5% fewer thinking tokens out of Qwen3.8-Flash-Next at ~1pp accuracy cost — Thom_Wolf · 2026-10-23
- DLOSAC refines RANSAC with Delaunay-guided spatial-geometric losses for geometry estimation — zhenjun_zhao · 2026-10-23
- New Benchmark Tests 3D Foundation Models on Geographically Diverse Architecture — zhenjun_zhao · 2026-10-23
- FineSplat: Fine-Grained Patches Push Feed-Forward 3D Gaussian Splatting to New SOTA — zhenjun_zhao · 2026-10-23
- ReDT: Resolution-Decoupled Transformer Brings 16x Speedup to HD 3D Reconstruction — zhenjun_zhao · 2026-10-23
- Causal Spatiotemporal Mamba Enables Feed-Forward Monocular 4D Reconstruction — zhenjun_zhao · 2026-10-23