**Experimental**: This documentation covers the AWS SDK for Python v4, which is experimental, non-GA software intended for evaluation and testing only. Do not use it for production workloads. For production applications, use the [AWS SDK for Python (Boto3)](https://docs.aws.amazon.com/boto3/latest/guide/quickstart.html). To understand the differences between the two SDKs, see [Choosing the right AWS SDK for Python](choosing-sdk.md).

# Key differences
<a name="working-with-boto3-differences"></a>

AWS SDK for Python (Boto3) and the AWS SDK for Python v4 have different execution models and Python client interfaces. The following table summarizes the API-model and execution differences that affect your application code. The operation, request, response, streaming, and error rows compare generated AWS SDK for Python v4 clients with Boto3 low-level clients.


| Area | AWS SDK for Python (Boto3) | AWS SDK for Python v4 | 
| --- | --- | --- | 
| Execution model | Service operations are synchronous and block the calling thread until they complete. | Service operations are asynchronous and yield to the event loop instead of blocking, so other tasks can run while a call is in flight. | 
| Packages and service coverage | Installing boto3 provides access to clients and features for a broad set of AWS services. | Each service client is distributed as a separate package. The experimental SDK covers a growing subset of AWS services, and supported clients do not yet implement every Boto3 feature. | 
| Clients and imports | Import boto3 and create clients dynamically, such as boto3.client("dynamodb"). | Install a service package and import its generated client, such as AsyncDynamoDBClient from aws-sdk-dynamodb. | 
| Higher-level features | Provides higher-level abstractions for supported services, including Resources, paginator and waiter objects, and managed transfer utilities. | The experimental SDK focuses on generated service clients. It doesn't currently provide Boto3-equivalent Resources, paginator or waiter objects, or managed transfer utilities. | 
| Operation and member names | Low-level client operation methods use snake case, such as put\_object, but request parameter names preserve modeled API casing, such as Bucket and Key. | Operation methods and generated input and output model members use snake case, such as start\_stream\_transcription() and media\_sample\_rate\_hertz. | 
| Requests | Low-level clients accept modeled request parameters as keyword arguments. | Construct a generated input model and pass it as the operation's input parameter, either positionally or with input=. | 
| Responses | Low-level client operations generally return service responses as dictionaries. Read values by key and check whether optional keys are present. Higher-level Boto3 Resources and helpers have API-specific return types, such as resource objects, iterables, or None. | Most non-streaming operations return generated output models, while streaming operations return typed asynchronous stream wrappers. Read model attributes and check optional members for None. When a member is a union, narrow its generated variant before accessing its value. | 
| Streaming | Boto3 clients don't provide native asynchronous operations. Reading supported streaming response bodies or event streams uses synchronous APIs and can block the calling thread. | Supported streaming operations expose typed asynchronous event streams. Some operations provide bidirectional streams over HTTP/2. | 
| Errors | Client-side failures use Botocore exception classes. Service errors can be caught as ClientError, which exposes details through its response dictionary. | Modeled service errors use concrete generated exception classes, including errors raised while consuming a stream. Other SDK runtime exceptions can occur during configuration, request preparation, transport, or response processing. | 
| Configuration | boto3.client() uses the default Boto3 session. Create a boto3.Session or pass a Botocore Config to customize settings. | Generated clients use their default configuration. AsyncAwsConfig resolves common AWS settings, and generated service-specific configuration classes add service settings. To customize a client, call the service-specific class's inherited resolve() method, such as await AsyncDynamoDBConfig.resolve(...), and pass the result. |