AI Model2026
Accepts text and images and generates text for long-context and agentic workflows. A shared-weight looped decoder and two-stream mHC add computation depth without duplicating the repeated block's weights.
AI Video Papers2026
Applies text- or image-guided edits across long, multi-shot videos while keeping the change consistent across cuts. Its training pipeline creates 87K paired editing videos without manual annotation.
AI Model2026
A multimodal model for coding, visual understanding, and multi-step tool use, with a 1-million-token context window. Its MoE architecture has 2.8T total parameters and activates 104B per token; weights are open, but local inference demands substantial compute.
AI Model2026
Compresses the model by pruning half its experts and quantizing the retained weights, targeting code, agentic workflows, long context, and vision. It retains 94.8%–98.7% of the base model’s scores on two coding benchmarks, at the cost of broader capabilities.
AI Video Papers2026
Improves when streaming video models respond or keep observing by assigning reinforcement-learning credit to individual turns. Its prefix-scoring and turn-mean objectives reduce redundant replies without a learned critic or extra rollouts.
AI Dataset2026
Converts 1,033 Google Scanned Objects household scans into web-ready GLB files, with KTX2 textures, LODs, and collision hulls for 3D scenes and simulation.
AI Dataset2026
Provides 7,953 browser-optimized GLB models of real retail products, with PBR materials, product metadata, and multiple detail levels for 3D rendering and real-world object understanding research.
Lets robot manipulation policies draw on visual representations from different depths of a vision-language backbone instead of relying on a separate encoder’s fixed output. The approach improves control results on LIBERO and real-world tasks.