Benchmark RadarAI BENCHMARK PROFILE
ReflexBench
Evaluates vision-language-action models on reaction-critical manipulation across six dynamic tasks, with configurable latency under synchronous and asynchronous inference in a simulated environment.
- Released
- 2026-08-14
- Readiness
- Inspectable
- Primary field
- Robotics & Autonomous Systems
Why it matters
Addresses the lack of benchmarks for dynamic interaction scenarios in robotic manipulation, providing a standardized way to assess model performance under reaction-critical conditions.
Motivation
Vision-Language-Action (VLA) models have recently achieved promising performance in robotic manipulation.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.