Claude Platform on AWS occupies a spot the other third-party options don't: it is operated by Anthropic itself but runs on AWS, inside the cloud boundary your organization already governs. The result is an unusual combination — typically same-day API parity with Anthropic's first-party API (a few admin and billing features are documented exceptions), plus AWS-native authentication. If your company is on AWS and wants new Claude capabilities the day they ship, this is the platform built for that. Here is the path to a first response.
How it differs from Bedrock in one minute
Both run on AWS, which invites confusion. Amazon Bedrock is an AWS-operated managed service: AWS runs it, AWS decides when new Claude features arrive, and some (Batch API, Files API, code execution, web tools) haven't arrived at all. Claude Platform on AWS is Anthropic-operated: features land the same day they ship on the first-party API, and it is the only 3P platform with Managed Agents. You keep AWS-style authentication and residency either way; what changes is who operates the service and how fast the feature set moves.
Prerequisites: credentials, a workspace, two environment variables
Working AWS credentials. Requests are signed with AWS SigV4 — the same request-signing mechanism every AWS SDK uses. You never handle the signing yourself; the client does it using whatever AWS credentials your environment already provides (environment variables, a credentials file, or an assumed role). If aws sts get-caller-identity works, this part is done.
A workspace ID. Claude Platform on AWS organizes usage into workspaces — a scoping unit that keeps teams, environments, or projects separate. Your administrator creates the workspace and gives you its ID.
Two environment variables. The client reads its configuration from AWS_REGION and ANTHROPIC_AWS_WORKSPACE_ID. Set both and the constructor needs no arguments at all.
Install the SDK and make the call
The platform uses Anthropic's official Python package:
pip install -U anthropic
import os
from anthropic import AnthropicAWS
# Reads AWS_REGION and ANTHROPIC_AWS_WORKSPACE_ID from the environment;
# requests are signed with your standard AWS credentials (SigV4).
client = AnthropicAWS()
message = client.messages.create(
model="claude-sonnet-5",
max_tokens=512,
messages=[
{"role": "user", "content": "In one sentence, what is SigV4 signing?"}
],
)
print(message.content[0].text)Model IDs are the bare first-party form — claude-sonnet-5, claude-opus-4-8, claude-haiku-4-5. Only Bedrock prefixes IDs with anthropic.; this platform does not, which also means code written here matches first-party code almost exactly.
ANTHROPIC_AWS_WORKSPACE_ID is the most common first-run error — then confirm your AWS credentials resolve to an identity your administrator actually granted access to the workspace.What you get for choosing this door
Everything, immediately. Core features (Messages API, streaming, tool use, vision, extended and adaptive thinking, prompt caching) plus the ones missing elsewhere: Batch API, Files API, code execution tool, web fetch and web search tools, and Managed Agents — the last available only here and on the first-party API. Same-day parity means your team never reads a release announcement and then waits.
AWS-shaped operations. SigV4 auth means no API keys to issue or rotate for this service, and the AWS-resident deployment fits cloud-boundary requirements many security teams already have. As with any platform choice, compliance specifics depend on your obligations — confirm them with your provider rather than assuming inheritance covers everything.
The trade-off is organizational: you add an Anthropic-operated service to an AWS estate, so your support and operational relationship spans two parties instead of one. For teams that value feature velocity, that is usually a fair trade.
A first-week habit worth keeping from the very first script: log the token usage figures each response reports. Input and output token counts are the units your spend is measured in, and because workspaces already separate teams and environments, pairing those counts with a workspace gives you clean per-team cost attribution before you have built any dedicated tooling. It is the cheapest piece of financial hygiene in this whole setup, and the hardest to retrofit once traffic is real.
Where to go next
Read workspaces and SigV4 authentication for the details behind those two environment variables, and why same-day parity matters for the strategic case. The quickstart compares all four platforms.