Evaluates whether vision–language models can convert 2D observations into precise 3D control by introducing RoTA-Bench, an interactive projection-alignment benchmark that synthesizes silhouette-preserving 3D objects, guarantees solvable initializations, and provides 674 tasks with a stratified 100-task test set.
Restores recoverability in on-policy distillation by having the teacher generate critiques of incorrect student rollouts, guide a critique-free student rewrite, and score the revised solution as privileged context; yields improved math-reasoning performance, largest gains on competition problems.
Uses an intermediate teacher checkpoint to gate transfer of RL-induced policy shifts in on-policy distillation by checking directional agreement between early and late teacher changes; suppresses endpoint shifts that reverse and requires one extra frozen-teacher scoring pass.
AI Model2026
Multimodal Mixture-of-Experts foundation model for long-context agent workloads — supports up to 1M-token contexts and uses aggressive KV-cache compression (FP4 + cross-layer reuse) to reduce memory and storage footprint while improving reasoning, coding and vision benchmark performance.