# Benchmarks { #benchmarks }

Independent TechEmpower benchmarks show **FastAPI** applications running under Uvicorn as [one of the fastest Python frameworks available](https://www.techempower.com/benchmarks/#section=test&runid=7464e520-0dc2-473d-bd34-dbdfd7e85911&hw=ph&test=query&l=zijzen-7), only below Starlette and Uvicorn themselves (used internally by FastAPI).

But when checking benchmarks and comparisons you should keep the following in mind.

### Hey API { #hey-api }

Once we have a FastAPI app with the models, we can use Hey API to generate a TypeScript client. The fastest way to do that is via npx.

```sh
npx @hey-api/openapi-ts -i http://localhost:8000/openapi.json -o src/client
```

This will generate a TypeScript SDK in `./src/client`.

The key features are:

* **Fast**: Very high performance, on par with **NodeJS** and **Go** (thanks to Starlette and Pydantic). [One of the fastest Python frameworks available](#performance).
* **Fast to code**: Increase the speed to develop features by about 200% to 300%. *
* **Fewer bugs**: Reduce about 40% of human (developer) induced errors. *

///

/// check | Inspired **FastAPI** to

Find a way to have a crazy performance.

That's why **FastAPI** is based on Starlette, as it is the fastest framework available (tested by third-party benchmarks).

///

### [Falcon](https://falconframework.org/) { #falcon }

The key features are:

* **Fast**: Very high performance, on par with **NodeJS** and **Go** (thanks to Starlette and Pydantic). [One of the fastest Python frameworks available](#performance).
* **Fast to code**: Increase the speed to develop features by about 200% to 300%. *
* **Fewer bugs**: Reduce about 40% of human (developer) induced errors. *

With **FastAPI** you get all of **Starlette**'s features (as FastAPI is just Starlette on steroids):

* Seriously impressive performance. It is [one of the fastest Python frameworks available, on par with **NodeJS** and **Go**](https://github.com/encode/starlette#performance).
* **WebSocket** support.
* In-process background tasks.
* Startup and shutdown events.

      /*
       * The following static final fields exist for performance reasons.
       *
       * In UnsignedBytesBenchmark, accessing the following objects via static final fields is the
       * fastest (more than twice as fast as the Java implementation, vs ~1.5x with non-final static
       * fields, on x86_32) under the Hotspot server compiler. The reason is obviously that the

- For relative URLs, only translate the link text. Do not translate the URL or its parts.

Example:

Source (English):

```
[One of the fastest Python frameworks available](#performance)
```

Result (German):

```
[Eines der schnellsten verfügbaren Python-Frameworks](#performance)
```

* Introducing `tf.types.experimental.AtomicFunction` as the fastest way to perform TF computations in Python.

    * Can be accessed through `inference_fn` property of `ConcreteFunction`s
    * Does not support gradients.

produce // non-canonical representations. fe.reduce() r.reduce() return fe == r } if err := quick.Check(f2, nil); err != nil { t.Errorf("failed FE->bytes->FE round-trip: %v", err) } // Check some fixed vectors from dalek type feRTTest struct { fe Element b []byte } var tests = []feRTTest{ { fe: Element{358744748052810, 1691584618240980, 977650209285361, 1429865912637724, 560044844278676}, b: []byte{74, 209, 69, 197, 70, 70, 161, 222, 56, 226, 229, 19, 112, 60, 25, 92, 187, 74, 222, 56, 50, 153, 51, 233,...