---
title: "Benchmarks"
canonical: https://primateintelligence.ai/technology/benchmarks
site: Primate Intelligence
---

# Benchmarks

What we measure (action detection, temporal localization, question-answering with timestamps),
against which baselines, and honest failure modes.

Darwin beats V-JEPA 2.1 on SSv2 and EK-100 action benchmarks at a fraction of the training cost.

- [Darwin: Video JEPA model that outperforms SOTA models while running on edge CPU](https://primateintelligence.ai/blog/darwin-video-jepa-benchmarks)
- [Benchmark post — methodology and extended results](https://primateintelligence.ai/blog/benchmark-post)