Research, releases, and field notes.
Open benchmark
robot-eval-bench
Product · Aug 14, 2026
Introducing Kite Evals
Evaluation is the slowest part of building a robot. We made it the fastest — a thousand scenarios in under an hour, with failures that explain themselves.
Read article
Latest
Two new judges on robot-eval-bench: Gemini 3.7 flash and Muse Spark 1.1
Research·Aug 13, 2026
Cornucopia Bio ships lab automation policies in hours, not weeks
Customer Stories·Aug 10, 2026
Robot-eval-bench: Benchmarking task success evaluation for VLMs
Research·Jul 30, 2026
Introducing the Kite Platform API
Product·Jul 29, 2026
The evidence for data augmentation in robotics
Research·Jul 28, 2026
Kite joins the NVIDIA Inception Program
Company·Jun 10, 2026