API compatibility - 3 documented limits
Amazon Bedrock API compatibility
Amazon Bedrock documents an OpenAI-compatible API at https://bedrock-runtime.{region}.amazonaws.com/openai/v1 (chat/completions, responses and models (bedrock-mantle only)). Its Anthropic-compatible base URL, the value for ANTHROPIC_BASE_URL, is https://bedrock-runtime.{region}.amazonaws.com/anthropic, and it serves Claude models only. The docs list 3 limitations: 3 for the OpenAI-compatible layer.
Verified against Amazon Bedrock's own documentation. Not documented means the docs read that day do not say; it never means No.
Compare every provider in the API compatibility matrix.
OpenAI-compatible API
Supportedhttps://bedrock-runtime.{region}.amazonaws.com/openai/v1- Endpoints
- chat/completions
- responses
- models (bedrock-mantle only)
- Auth
- Authorization: Bearer <Amazon Bedrock API key>, or AWS SigV4 credentials
- Regional and alternate URLs
- Bedrock Runtime
https:// bedrock-runtime. {region}. amazonaws. com/ openai/ v1 - Bedrock Mantle
https:// bedrock-mantle. {region}. api. aws/ v1
- Bedrock Runtime
- Notes
- bedrock-runtime is recommended for new applications; bedrock-mantle is a compatibility option when a model or capability is not on bedrock-runtime. On bedrock-mantle some models are served under /openai/v1 instead of /v1; check each model card. Per-model API support varies (for example GPT OSS supports Chat Completions but not Responses on bedrock-runtime). The base URL is per region.
From the docs
bedrock-runtime(recommended) |https://bedrock-runtime.{region}.amazonaws.com/openai/v1/chat/completions| AWS credentials (SigV4) or Amazon Bedrock API key ...bedrock-mantle(compatibility) |https://bedrock-mantle.{region}.api.aws/v1/chat/completions| Amazon Bedrock API key or AWS credentials
Anthropic-compatible API
SupportedClaude models only. This endpoint gives native Messages API access to Anthropic's Claude models; it does not put other model families behind the Anthropic format.
https://bedrock-runtime.{region}.amazonaws.com/anthropic- Auth
- x-api-key: <Amazon Bedrock API key> (docs use $AWS_BEARER_TOKEN_BEDROCK / BEDROCK_API_KEY), a short-term Bedrock bearer token, or AWS SigV4
- Regional and alternate URLs
- Bedrock Runtime
https:// bedrock-runtime. {region}. amazonaws. com/ anthropic - Bedrock Mantle
https:// bedrock-mantle. {region}. api. aws/ anthropic
- Bedrock Runtime
- Notes
- This is native Anthropic Messages API access to Claude models on Bedrock, not a translation layer for other model families. bedrock-mantle requires the anthropic-version: 2023-06-01 header; count_tokens is on bedrock-mantle.
From the docs
The Anthropic Messages API provides native access to Claude models using the Anthropic request and response format. You can use the Messages API on both thebedrock-runtimeandbedrock-mantleendpoints. For new applications, we recommend thebedrock-runtimeendpoint.
Documented limitations
Parameters, content types and features Amazon Bedrock's docs say its compatibility layers ignore, reject or change, each with the line it comes from. A layer with nothing listed may still differ in ways the docs do not spell out.
OpenAI-compatible layer3
GET /models not implemented on bedrock-runtime
bedrock-runtimedoesn't implement the OpenAI-compatibleGET /modelsoperation, soclient.models.list()andGET /openai/v1/modelsdon't work on this endpoint.docs.aws.amazon.com/bedrock/latest/userguide/inference-chat-completions.html GPT OSS models: no Responses API on bedrock-runtime
GPT OSS models support Chat Completions on
bedrock-runtime, but don't support the Responses API there.docs.aws.amazon.com/bedrock/latest/userguide/inference-chat-completions.html bedrock-mantle path prefix varies by model
On
bedrock-mantle, the path prefix varies by model. Most models are served under/v1, as shown in the preceding table, but some are served under/openai/v1instead.docs.aws.amazon.com/bedrock/latest/userguide/inference-chat-completions.html
Anthropic-compatible layer0
No limitations listed in the docs read.
Example from Amazon Bedrock's docs
Copied verbatim from the provider's documentation, placeholders included.
from openai import OpenAI
import os
client = OpenAI(
base_url="https://bedrock-runtime.us-east-1.amazonaws.com/openai/v1",
api_key=os.environ.get("OPENAI_API_KEY")
)
response = client.chat.completions.create(
model="openai.gpt-oss-120b-1:0",
messages=[{"role": "user", "content": "Hello"}]
)More on Amazon Bedrock
Newsletter
Know when a provider changes its API
One email per week. New models, new endpoints and the compatibility changes that break a base URL, and what they mean for your stack.