Captures synchronized multimodal embodied-human data in real homes — egocentric and multi-view video, metric body/hand/object motion, audio, and tactile signals. Released under a gated non-commercial research license with identifiable participants and strict non-redistribution/privacy constraints.
Generated instance set (seed 31415) for ASI‑Bench: includes four matched prompt variants, agent-visible inputs, reference artifacts, and instance metadata for 60 project-scale scientific research tasks across 11 domains; intended for evaluating autonomous research agents. Licensed Apache‑2.0.
An end-to-end 11B full-duplex speech model for real-time conversational AI that jointly performs streaming speech understanding and generation, enabling ~450 ms turn-taking, barge‑in and live tool calling in a single unified architecture; research use only.
Estimates the visually attributable portion of a privileged teacher’s next-token corrections and reconstructs student-anchored training targets for multimodal on-policy distillation. Uses counterfactual teacher queries and a signed proxy to raise supported tokens and suppress refuted ones, improving fine-grained visual knowledge transfer across model scales.
Evaluates vision-language model judges on computer-using agent (CUA) trajectories to measure verifier reliability. Provides OSReward-Hard and OSReward-Multi challenge sets, the OS-Shepherd-100K reasoning-annotated corpus, and trained OS-Shepherd reward models that match commercial judges at ~30–60× lower cost.
A 365-day, order-level simulation benchmark for evaluating long-term coherence of LLM agents in seller-side e-commerce. Grounded in 98,843 real product records and 26 interactive tools, it pairs prompt upstream supplier signals with delayed downstream order outcomes to stress planning, memory, and tool use over long horizons.
Evolves persistent, stateful environments to red-team tool-using AI agents — provides 10K+ validated scenarios across 50 domains and a feedback-driven attack policy (EMHA) to surface long‑horizon safety failures.
Introduces WorldExam, a diagnostic benchmark that evaluates controllable video world models across four levels from visual quality to inherent world reactivity. Covers 1,474 cases across eight tasks and supports camera-, action-, and language-driven paradigms, measuring scene-conditioned reactions beyond explicit instructions.
Performs real-time, instruction-guided video-to-video editing on streaming input using a 16B autoregressive diffusion model that preserves subject identity and long-term temporal coherence; achieves end-to-end 720p at ≈30 FPS on a single Nvidia B200 GPU. Key features include chunk-wise autoregressive adaptation, Source-Anchored Distribution Matching Distillation (SA-DMD) that reduces diffusion to a two-step generator, and Long-Horizon Autoregressive Distillation to mitigate temporal drift.
Analyzes how to build effective training environment distributions for multimodal agents and proposes Ability-aware Environment Selection (AES) and Hierarchical Difficulty Curriculum (HDC) to improve diversity and difficulty scheduling, yielding large relative gains in experiments.
Evaluates VLMs' ability to form global spatial awareness from long-horizon egocentric video. Introduces GST-Bench: a VQA benchmark with human-verified questions from 6,790 minutes of synthetic video, reveals a large gap (best zero-shot 42.68 vs human 79.08) and provides GST-Train dataset.
Evaluates schema-guided structured extraction from documents: given a document and a JSON schema, systems must return a schema-valid JSON with page-and-box grounding. Covers 370 documents (4,869 pages) across 8 business domains and 67 document types; scores value accuracy, word/page grounding, and long-list completeness.