Skip to content

Eureka

Reward design through coding language models

Visit official site

What Eureka is

Eureka is described as a reward design algorithm implemented by coding large language models that evolve reward code for reinforcement learning.

What you can do with it

  • Design reward functions with an LLM
  • Evolve complex reward code
  • Use generated rewards for RL training

Model notes

  • The material lists 29 open RL environments
  • Ten robot forms are named in the results
  • A pen-spinning Shadow Hand demo is described

Pros & cons

✓ What we like

  • The material reports results across 29 RL environments
  • Ten different robot forms are named

! What to watch out for

  • The approach relies on a coding language model
  • The listed output is reward code for RL

FAQ

Which model is named?

The material names GPT-4 among the language models used.

What skill is demonstrated?

The material describes a simulated Shadow Hand spinning a pen.

Last reviewed: 2026-11-15

More AI coding tools tools

View all →

How we review