---
format: "aidr-story-markdown/v1"
id: "01b8bd072db9d88b3991380e19eadcf63b7629eb9efe119ea1769cc8f338c013"
canonical_url: "https://aidr.today/01b8bd07?lang=en"
title: "LeVJEPA Matches V-JEPA 2 With 20x Less Video Pretraining Compute"
lang: "en"
requested_lang: "en"
available_langs: ["en","vi"]
translation_fallback: null
fallback_fields: []
published_at: "2026-08-29T05:37:48.000Z"
category: "Research"
topics: ["open-source","inference","video"]
source_urls: ["https://huggingnews.com/ai/levjepa-matches-v-jepa-2-with-20x-less-video-pretraining-compute-45862a44","https://x.com/lukaskuhn77/status/2093318310779613563","https://x.com/randall_balestr/status/2093322239529652671","https://x.com/askalphaxiv/status/2093525041296396351","https://x.com/ylecun/status/2093549515831701539"]
summary: "The new pretraining method replaces target encoders and reconstruction machinery with a single encoder trained using invariance and SIGReg. This end-to-end approach achieves performance equivalent to V-JEPA 2 while reducing compute costs by 5.6 to 20.8 times. LeVJEPA utilizes a stable recipe to maintain appearance and motion understanding, even when randomly dropping 95% of video tokens. The method is open source and removes the need for masked predictions, stop-gradients, or teacher-student schedules."
---

# LeVJEPA Matches V\-JEPA 2 With 20x Less Video Pretraining Compute

> [Open the canonical story](<https://aidr.today/01b8bd07?lang=en>)

**Published:** 2026-08-29T05:37:48.000Z
**Category:** Research
**Topics:** open\-source, inference, video

## Summary

The new pretraining method replaces target encoders and reconstruction machinery with a single encoder trained using invariance and SIGReg\. This end\-to\-end approach achieves performance equivalent to V\-JEPA 2 while reducing compute costs by 5\.6 to 20\.8 times\. LeVJEPA utilizes a stable recipe to maintain appearance and motion understanding, even when randomly dropping 95% of video tokens\. The method is open source and removes the need for masked predictions, stop\-gradients, or teacher\-student schedules\.

## Sources

- [Story source](<https://huggingnews.com/ai/levjepa-matches-v-jepa-2-with-20x-less-video-pretraining-compute-45862a44>)
- [Story source](<https://x.com/lukaskuhn77/status/2093318310779613563>)
- [Supporting source](<https://x.com/randall_balestr/status/2093322239529652671>)
- [Supporting source](<https://x.com/askalphaxiv/status/2093525041296396351>)
- [Supporting source](<https://x.com/ylecun/status/2093549515831701539>)