Website
A dedicated home for benchmark results.
Leaderboard and per-task results now share this page, with capability, metric, and domain filters. Explore the rankings →
← Back to the benchmark
BiDex-100 · Evaluation
Benchmark results.
Compare policies across capabilities.
Then look closer at every task.
7 VLA models106 tasksMatched ID / OOD100 rollouts per task · 50 ID + 50 OOD
News
Updates from the benchmark.
Task gallery
Full-length videos for all 106 tasks.
The task gallery now includes complete demonstrations and rollouts, replacing short previews. Watch the replays →
Evaluation
Seven-model ID / OOD results updated.
Aggregate rankings, capability breakdowns, and individual task results reflect the evaluation snapshot dated October 19. Inspect detailed results →
Leaderboard
7 co-trained models.
SR strict success rate · Score progress score.
Every result cell is shown as SR / Score. Ranking follows the selected dimension, metric, and domain; capability columns are equal-weight means over their tasks. Best values are highlighted.
Detailed Results
Per-task evaluation results · 106 tasks × 7 models.
Each task has 100 rollouts in total: 50 In-Domain + 50 Out-of-Domain (2 fixed seeds × 25 episodes per domain). Strict success is shown first by default; toggle the metric or domain to inspect partial progress and OOD degradation.
Metric
Domain
Cell shading scales with the value (0–100). Rows are grouped by capability axis (colored header); the Overall mean footer is the equal-weight per-model mean over 212 task-runs (106 tasks × 2 domains). Select a task name for its full replay and per-subtask breakdown. Best value per task is bolded.