Every RL environment, from every hub.
Environments, benchmarks, datasets, agents and verifiers indexed from the hubs that already exist, with attribution, and with a measured tier instead of a claim.
Search it like a library.
An RL environment is a task a model practises on, with a scorer that says how well it did. Explore puts every one we can find in one search.
- Filter by what you needKind, source hub, domain (code, reasoning, games…) and conformance tier.
- Attribution keptEach listing keeps its owner, its tags and the hub it came from.
- One click to use itOpen a listing and use it in a project to evaluate or train on your own stack.
open-thought
basic_arithmetic
Procedurally generated arithmetic with a built-in verifier.
Audited
harbor-framework
terminal-bench
Agents solving tasks in real terminal environments.
Not checked
primeintellect
wordle
Wordle game environment.
Not checked
Farama Foundation
CartPole-v1
The classic cart-pole control problem.
Not checked
Know what you're training on.
Most hubs list environments as if they all work. Nenyax measures them, and keeps problems in the open.
nenyax checkreasoning_gym:basic_arithmetic
nenyax checkreasoning_gym:basic_arithmetic
Browse the way you think.
Benchmarks
One benchmark across every hub: its copies side by side, the one to use, and the best measured score.
Tasks
Browse by what you want to train: math, code, games, the web and more.
Organizations
Labs, companies and hub publishers, with everything each one has published.
Indexed from 13 hubs.
Ready hubs are indexed and searchable today. Beta hubs pull models and datasets.
Prime Intellect HubReady
Hugging FaceReady
OpenAI EvalsReady
lm-eval-harnessReadyreasoning-gymReadyHarborReadyGymnasiumReady
GitHubReady
Gitee AIBeta
KaggleBeta
ModelersBeta
NVIDIA NGCBeta
OpenXLabBetaMore than a catalogue.
The hub is where people compare, report and reward work on environments.
Three kinds of bountyreward in US dollars
Rewards for results
Post a reward for a score, a new environment or a fix. Score bounties only accept runs that reach the target.
Kinds
Not just environments
Datasets, agents, algorithms and verifiers sit in the same search.
Head to head
Models compete on the same games, under the same rules.
Your listingmaya / support-inbox
Your work, under your name
Push an environment or benchmark, pass the audit, and make it public when you're ready.
Find it, check it, run it.
Find
Search by kind, source, domain or tier.
Check
Read its tier, its scorer and any advisories.
Use
Add it to a project with the model you want to measure.
Run
Your runner executes it and the score lands on the leaderboard.
Good to know.
Find the environment you need.
Search the whole index, check the tier, and run it on your own stack.