← Back to Videos

LeVJEPA 1.5 Scales Video Learning With Far Less Compute

Loading video

LeVJEPA 1.5 Large closes much of the gap to V-JEPA on image and video benchmarks while using 5.6 times less compute and over an order of magnitude less data.

JEPA Video Representation Self-Supervised Learning LeVJEPA

Category: research

Author: @lukaskuhn77

Date: 2026-11-02T00:00:00

Duration: 34.5s

Source: https://x.com/lukaskuhn77/status/2117323261465735320

RoboJEPA: The First Scaling Law for Multi-Embodiment World Models

RoboJEPA: The First Scaling Law for Multi-Embodiment World Models

@artemZholus · 315 views · 2026-10-10

World ModelsJEPA

JEPA-Anything: One World Model Across Six Domains

JEPA-Anything: One World Model Across Six Domains

@Charles_Y_Wu · 477 views · 2026-10-05

World ModelsEmbodied AI

T2-z Adapts World Models at Test Time

T2-z Adapts World Models at Test Time

@yining_hong · 18 views · 2026-10-16

World ModelsTest-Time Optimization

1.5M-Splat Cherry Blossom Street Reconstruction

1.5M-Splat Cherry Blossom Street Reconstruction

@metaseq3d · 35 views · 2026-11-02

3D Gaussian SplattingStructure-from-Motion

Robot Simulation Needs Accurate Physics, Not Just Visuals

Robot Simulation Needs Accurate Physics, Not Just Visuals

@PranjalAggarw16 · 48 views · 2026-11-02

Robotics SimulationPhysics

DeepSeek Harness Opens Creator Mode for Plugins

DeepSeek Harness Opens Creator Mode for Plugins

@DeepSeekHarness · 49 views · 2026-11-02

Agent HarnessPlugins

As an Amazon Associate, we earn from qualifying purchases.