# 600 JEPA world-model experiments — subgoals, not physics, limit planning

> A long hands-on study finds that a JEPA world model's planner, not its predictor, is the real bottleneck.

- Org: Haddy Alchaer
- Date: 2026-11-11
- Category: article, algorithm
- Source: https://haddyalchaer.com/jepa
- Canonical page: https://ai-tldr.dev/releases/haddy-alchaer-jepa-world-models-600-experiments/

## In short

Haddy Alchaer ran about 600 experiments with the LeWM JEPA world model on Push-T. Its learned predictor plans as well as true physics, while adding subgoals lifts long-horizon success from 22% to 80%.

## What it is

Haddy Alchaer's post reports on about 600 experiments with LeWM, a JEPA world model from Yann LeCun's group that predicts the future as embeddings instead of pixels. All tests use the Push-T robot task and the published model, with no retraining.

## How it works

On 250 tasks with a goal 75 steps away, the learned predictor succeeded 78% of the time against 77% for the simulator's exact physics. Planning all 75 steps at once gave only 22%; a subgoal every 25 steps raised it to 80%. A memory graph built from about 460,000 embedded demonstration states, with a cost on each jump, reached 85%.

## Why it matters

The study points researchers at planning and goal choice, not better predictors, as the next problem for JEPA-style world models. It also shows the standard test is easy: LeWM scored 68% on the official 50 episodes but 44% on 250 fresh ones, so reported Push-T numbers may be inflated.

## Who it's for

world-model and robot-learning researchers

## Links

- [Blog Post](https://haddyalchaer.com/jepa)
- [LeWM Paper](https://arxiv.org/abs/2603.19312)
- [HN Discussion](https://news.ycombinator.com/item?id=50487185)