Robocurve releases all 2,109 robot runs with transcripts and videos; Gemini 4 Argon averages 33.8%
zacharynado · x · 2026-10-15
Robocurve has released its entire dataset of 2,109 robot runs with full transcripts, videos, and downloadable CSVs, covering benchmarks like RoboDojo-RC Tier 1 and RoboHarm v1 across multiple frontier models.
- The 120 released Gemini 4 Argon runs average 33.8% on RoboDojo-RC Tier 1: Cap Pen 42.9%, Sort Objects 37.1%, Pack And Pour Fruit just 23.4%
- Runs are filterable by benchmark, task, model, scene (e.g., wrapped vs unwrapped baby doll), thinking budget, and control mode
- Outcomes include safety guardrail blocks and harmful-instruction refusals, with estimated cost and call counts per run
A rare fully transparent release of robot model evals, enabling community reproduction and head-to-head comparison of frontier models on real robot tasks.
More from Embodied
- Autonomous sells personal AI computers from $26,100 that keep data in the room — dee_hw · 2026-10-15
- Bee launches Frontier and Eclipse, two wearable AI memory devices — Scobleizer · 2026-10-15
- How Microsoft optimized Windows for Nvidia's Arm-based RTX Spark laptops — The Verge AI · 2026-10-15
- John Deere's second-gen autonomous tillage: 16 cross-checking cameras beat sensor drift — lukas_m_ziegler · 2026-10-15
- Europe's Application-Layer Bet: Can the Rotterdam Humanoid Center Deliver? — ingliguori · 2026-10-15
- VC thesis 'Data at the Edge': cheap sensors plus AI are rebuilding the internet's data flywheel offline — rebeccakaden · 2026-10-15