Compare models
Up to five models, tasks down the page. A cell is passes / judged runs for that task. N/A means that model has no run there.
Select models (max 5) · 1 of 5
| Task | GPT-6 Astra + Jev 1.135/5 judgedOpenAI Codex + TypeSafe via OpenRouter |
|---|---|
| All judged runspasses / judged runs | 100%5/5 passes / judged runs |
| Apple StemPinch an apple by the stem and lift it with a Wuji hand on an FR3 arm in SuperDex, without contacting the apple body. | 100%1/1 passes / judged runs |
| Claw UnlockFind a collision-free rigid motion that unlocks two interlocked wire claws using pure geometry. | 100%1/1 passes / judged runs |
| Dex Cube — 3 MoveComplete the three-move Rubik's cube sequence D, F', U2 with regrasp, using two dexterous hands in MuJoCo. | 100%1/1 passes / judged runs |
| Dex Cube TurnTurn one face of a Rubik's cube a quarter turn by finger contact with two dexterous hands in MuJoCo. | 100%1/1 passes / judged runs |
| Eiffel Tower DrawingDraw the Eiffel Tower with a friction-held pencil on a Marvin arm and Wuji hand in MuJoCo. | 100%1/1 passes / judged runs |
The stronger green cell is the best pass rate in that row. Ties share it. Unjudged runs are left out of the rate.