RLE-Bench2026
OpenFridge Kinex and Codex Evaluation
Ten recorded OpenFridge episodes evaluating GPT-6 Astra with Kinex and Codex harnesses on refrigerator manipulation tasks. Includes full session transcripts, model records, tool call histories, workspace archives, and edited videos with failure/success analysis.
Downloads85
Episodes10
Why This Matters for Physical AI
Provides empirical evaluation data comparing different language model integration harnesses (Kinex vs Codex) for robotic manipulation in simulation, demonstrating the impact of model-loop constraints and reflection mechanisms on task success rates.
Technical Profile
- Modalities
- rgbproprioceptionlanguage
- Action Space
- control_commands
- Environment
- simulation
- Task Types
- manipulationdoor_opening
- Episodes
- 10
- Data Format
- native_format
- Annotation Types
- language_instructionsreward_labelsaction_labelsfailure_success_analysis
Access
Need custom rgb data?
Claru builds purpose-built datasets for simulation applications with dense human annotations and quality assurance.
Request a Sample Pack