Bench

Reactor Bench is in preview.

Enter the preview password to continue.

Bench
Reactor Bench

Run your world model through the Reactor Runtime on shared benchmarks, and see how it ranks against every other model.

View the leaderboards
1benchmark
2tracks
447cases
0model entries
How it works

From your model's code to the leaderboard.

01 · Port

Wrap your model in the Reactor Runtime.

Your model's code goes inside the runtime, which gives it one way in and one way out.

02 · Drive

The runtime drives it, case by case.

Each turn's input goes in and a frame comes out. The frames make the rollout.

03 · Score

The rollout is scored.

Every frame is measured against the benchmark's metrics, and the metrics add up to its dimensions.

04 · Rank

Your model takes its place.

Its scores put it on the leaderboard, ranked against every other model.

View the leaderboards
Agentic-ready

Your agent can do the whole thing.

Pick your agent and it opens with the task filled in: port your model, score it on a track, package the run and submit it. You read the prompt, and press Enter.

Works with Cursor, Codex, Claude Code, Claude, VS Code and Warp, through an open API.

Your agent opens with this prompt filled in.
Leaderboards

Your model, ranked on every benchmark.

Score it on a benchmark and see where it stands, next to every other model.

WBench

WBench Navi158 cases · 0 modelsNo models yet
WBench Full289 cases · 0 modelsNo models yet
View the leaderboard →
Coming soon

More benchmarks

New benchmarks join the list as they are added.

Request one →
Get ranked

Put your world model on the leaderboard.

View the leaderboards