diff --git a/scripts/benchmarks/README.md b/scripts/benchmarks/README.md
index 95e5045a7..a22f3c7ad 100644
--- a/scripts/benchmarks/README.md
+++ b/scripts/benchmarks/README.md
@@ -1,70 +1,7 @@
-# benchmarks
+# Benchmarks
 
-Try the [benchmarks app](https://necolas.github.io/react-native-web/benchmarks) online.
+These benchmarks test how fast various css-in-js libraries render with react components. They do not necessarily represent how fast a css-in-js library will be in the real world and are used by emotion to stop performance regressions in render performance/caching.
 
-To run the benchmarks locally:
+To run these benchmarks, visit https://benchmarks.emotion.sh and choose a commit.
 
-```
-yarn benchmarks
-open ./packages/benchmarks/dist/index.html
-```
-
-Develop against these benchmarks:
-
-```
-yarn compile --watch
-yarn benchmarks --watch
-```
-
-## Notes
-
-These benchmarks are approximations of extreme cases that libraries may
-encounter. Their purpose is to provide an early-warning signal for performance
-regressions. Each test report includes the mean and standard deviation of the
-timings, and approximations of the time spent in scripting (S) and layout (L).
-
-The components used in the render benchmarks are simple enough to be
-implemented by multiple UI or style libraries. The benchmark implementations
-and the features of the style libraries are _only approximately equivalent in
-functionality_.
-
-No benchmark will run for more than 20 seconds.
-
-### Mount deep/wide tree
-
-These cases look at the performance of mounting and rendering large trees of
-elements that use static styles.
-
-### Update dynamic styles
-
-This case looks at the performance of repeated style updates to a large mounted
-tree. Some libraries choose to inject new styles for each "dynamic style",
-whereas others choose to use inline styles. Libraries without built-in support
-for dynamic styles (i.e., they rely on user-authored inline styles) are not
-included.
-
-## Example results
-
-### MacBook Pro (2011)
-
-MacBook Pro (13-inch, Early 2011); 2.3 GHz Intel Core i5; 8 GB 1333 MHz DDR3 RAM. Google Chrome 63.
-
-Typical render timings: mean ± standard deviations.
-
-| Implementation                        | Mount deep tree (ms) | Mount wide tree (ms) | Dynamic update (ms) |
-| :--- | ---: | ---: | ---: |
-| `css-modules`                         |     `30.19` `±04.84` |     `38.25` `±04.85` |                   - |
-| `react-native-web@0.4.0`              |     `36.40` `±04.98` |     `51.28` `±05.58` |    `19.36` `±02.56` |
-| `inline-styles`                       |     `64.12` `±07.69` |     `94.49` `±11.34` |    `09.84` `±02.36` |
-
-### Moto G4
-
-Moto G4 (Android 7); Octa-core (4x1.5 GHz & 4x1.2 Ghz); 2 GB RAM. Google Chrome 63.
-
-Typical render timings: mean ± standard deviations.
-
-| Implementation                        | Mount deep tree (ms) | Mount wide tree (ms) | Dynamic update (ms) |
-| :--- | ---: | ---: | ---: |
-| `css-modules`                         |     `98.24` `±20.26` |    `143.75` `±25.50` |                   - |
-| `react-native-web@0.4.0`              |    `131.46` `±18.96` |    `174.70` `±14.88` |    `60.87` `±06.32` |
-| `inline-styles`                       |    `184.58` `±26.23` |    `273.86` `±26.23` |    `30.28` `±07.44` |
+These benchmarks are a modified version of react-native-web's benchmarks.