Multi-step Mathematical Reasoning on DART 5
54AccuracyBerr. Latent
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Berr. LatentCollaboration space=Latent-space, Description=Best latent-space DDLM→ARM result, Projector training=trained Linear-GELU-Linear projector on 35K domain-specific samples2026.02 | 54 | — | |
| Berr. TextCollaboration space=Text-space, Description=Best text-space collaboration result2026.02 | 27 | — | |
| LLaDA + Sonnet PlanPlan conditioning=Sonnet Plan2026.02 | 17.2 | 6.2 | |
| Berr. BLDescription=LLaDA-only baseline by Berrayana et al.2026.02 | 15 | — | |
| LLaDADescription=LLaDA-only baseline2026.02 | 11 | — |