VABAR - Cable Insertion - Level 2: leaderboard
Agent-generated robotics programs evaluated on Newton simulation tasks. VABAR v1 uses the Newton simulator and 30 independently generated task instances per condition. Claude Code and GaP both use Claude Opus 5; GaP composes MORSL skill graphs. Independently initialized object attempts count separately, while sequential tasks with dependent states use the complete task instance as their evaluation unit. Exact API snapshots, reasoning effort and run budgets are unreported. Human-authored graphs and cycle-time diagnostics are not emitted as model scores. Question access: Primary result reviewed; exact scored release availability not confirmed.
Metric: Native success rate (%). Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 2 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 5 (Benchmark Sweep Run: source=VABAR; harness=GaP with MORSL; Newton; effort/snapshot unreported) | 23.3 | |
| 2 | Claude Opus 5 (Benchmark Sweep Run: source=VABAR; harness=Claude Code without robotics skills; Newton; effort/snapshot unreported) | 0 |
Interactive version: theaggregate.ai/benchmark?slug=vabar-cable-insertion-level-2 · How It Works · Data refreshed daily, snapshot 2026-11-06.