guides/amazon-bedrock.md +232 −0 created
1# OpenAI models in Amazon Bedrock
2
3Amazon Bedrock makes supported OpenAI models available through AWS-managed
4infrastructure. This deployment path is useful when your organization wants to
5keep procurement, identity, regional controls, and related cloud operations in
6AWS.
7
8Amazon Bedrock availability differs from the OpenAI API. Confirm the supported
9 model, AWS Region, feature set, and pricing path for your workload before you
10 deploy.
11
12## How Bedrock availability works
13
14OpenAI models in Amazon Bedrock run through an AWS-managed deployment path with
15Responses API compatibility for supported models and capabilities.
16Your application still uses OpenAI model behavior, but AWS owns the surrounding
17cloud control plane, including account access, regional availability, and
18billing.
19
20Use Bedrock when you need:
21
22- AWS-native procurement and billing.
23- AWS-managed identity, access, and account controls.
24- Deployment in supported AWS Regions for customers with cloud-location
25 requirements.
26
27Use the OpenAI API directly when you need the broadest feature coverage, the
28latest first-party platform capabilities, or functionality unavailable in
29Bedrock.
30
31## Make Responses API requests
32
33To send OpenAI SDK requests through Amazon Bedrock, use the Bedrock-aware SDK
34client and select the AWS Region and model ID for your deployment:
35
36- Instantiate `BedrockOpenAI` instead of the default `OpenAI` client. The client
37 derives the regional Mantle base URL from the AWS Region.
38- For the initial `openai.gpt-5.5` deployment, use `us-east-2`. This resolves to
39 `https://bedrock-mantle.us-east-2.api.aws/openai/v1`.
40- Use a Bedrock model ID with the `openai.` prefix, such as
41 `openai.gpt-5.5`.
42
43This example uses `openai.gpt-5.5` in `us-east-2`. Use a supported model and AWS
44Region combination for your Bedrock deployment.
45
46The following example uses a Bedrock API key stored as
47`AWS_BEARER_TOKEN_BEDROCK`. See
48[Amazon Bedrock API keys](https://docs.aws.amazon.com/bedrock/latest/userguide/api-keys.html)
49for information about generating and using a Bedrock API key. The SDK reads the
50token from your environment.
51
52Send a Responses API request through Amazon Bedrock
53
54```javascript
55import { BedrockOpenAI } from "openai";
56
57const client = new BedrockOpenAI({
58 awsRegion: "us-east-2",
59});
60
61const response = await client.responses.create({
62 model: "openai.gpt-5.5",
63 input: "Write a haiku about cloud infrastructure.",
64});
65
66console.log(response.output_text);
67```
68
69```python
70from openai import BedrockOpenAI
71
72client = BedrockOpenAI(aws_region="us-east-2")
73
74response = client.responses.create(
75 model="openai.gpt-5.5",
76 input="Write a haiku about cloud infrastructure.",
77)
78
79print(response.output_text)
80```
81
82```bash
83curl "https://bedrock-mantle.us-east-2.api.aws/openai/v1/responses" \\
84 -H "Content-Type: application/json" \\
85 -H "Authorization: Bearer $AWS_BEARER_TOKEN_BEDROCK" \\
86 -d '{
87 "model": "openai.gpt-5.5",
88 "input": "Write a haiku about cloud infrastructure."
89 }'
90```
91
92
93For long-running applications, pass a token provider instead of a static API
94key. The SDK calls the provider before each request. The AWS token-generator
95packages return a cached short-term key when the current key is valid and
96generate a new key when needed. They use the AWS credential chain, which can
97include credentials configured with `aws login`.
98
99Install the token-generator package for your SDK:
100
101```shell
102npm install @aws/bedrock-token-generator
103pip install aws-bedrock-token-generator
104```
105
106Send a request with refreshable Bedrock credentials
107
108```javascript
109import { getTokenProvider } from "@aws/bedrock-token-generator";
110import { BedrockOpenAI } from "openai";
111
112const client = new BedrockOpenAI({
113 awsRegion: "us-east-2",
114 bedrockTokenProvider: getTokenProvider(),
115});
116
117const response = await client.responses.create({
118 model: "openai.gpt-5.5",
119 input: "Write a haiku about cloud infrastructure.",
120});
121
122console.log(response.output_text);
123```
124
125```python
126from aws_bedrock_token_generator import provide_token
127from openai import BedrockOpenAI
128
129client = BedrockOpenAI(
130 aws_region="us-east-2",
131 bedrock_token_provider=provide_token,
132)
133
134response = client.responses.create(
135 model="openai.gpt-5.5",
136 input="Write a haiku about cloud infrastructure.",
137)
138
139print(response.output_text)
140```
141
142
143## Availability and operations
144
145Availability depends on AWS Region and model. The initial launch scope is more
146limited than the OpenAI API, so check [model support by AWS
147Region](https://docs.aws.amazon.com/bedrock/latest/userguide/models-region-compatibility.html)
148before rollout.
149
150Amazon Bedrock provides Responses API-compatible inference for supported OpenAI
151models in supported AWS Regions in the United States. AWS manages authentication,
152account access, procurement, and billing.
153
154AWS Regions are physical deployment locations, which differ from OpenAI data
155residency jurisdictions. Teams with residency requirements should evaluate the
156Bedrock Region itself and the corresponding AWS terms.
157
158## Responses API feature availability
159
160Amazon Bedrock supports a subset of Responses API capabilities available
161through the OpenAI API. This table describes intended feature availability for
162the initial Amazon Bedrock offering. It excludes transient availability and
163service status.
164
165The information below represents feature availability as of June 1, 2026.
166 Model and Region availability can also change. For the latest information, see
167 the [AWS documentation for OpenAI models in Amazon
168 Bedrock](https://docs.aws.amazon.com/bedrock/latest/userguide/model-cards-openai.html)
169 and [model support by AWS
170 Region](https://docs.aws.amazon.com/bedrock/latest/userguide/models-region-compatibility.html).
171
172| Capability | OpenAI API | Amazon Bedrock at initial availability |
173| ------------------------- | ------------------------- | -------------------------------------- |
174| Text generation | Available | Available |
175| Audio input | Available | Not available |
176| Image input | Available | Available |
177| File input | Available | Available for supported file types |
178| Structured outputs | Available | Available |
179| Function calling | Available | Available |
180| Streaming responses | Available | Available |
181| WebSocket connections | Available | Not available |
182| Reasoning effort | Available | Available |
183| Prompt caching | Available | Available |
184| Custom tools | Available | Available |
185| Client-side `tool_search` | Available | Available |
186| Hosted web search | Available | Not available |
187| Hosted file search | Available | Not available |
188| Computer use | Available | Not available |
189| Shell tool | Available | Not available |
190| Image generation tool | Available | Not available |
191| Remote MCP servers | Available | Not available |
192| Service tiers | Available where supported | On-demand inference only |
193
194Client-side `tool_search` is distinct from hosted tools and remote MCP server
195support. Hosted tools run through OpenAI-operated service infrastructure and
196are unavailable in the initial Amazon Bedrock offering.
197
198Treat feature parity as workload-specific. If your application depends on a
199specific tool, response mode, or service tier, test that behavior through
200Bedrock before you commit to the deployment path.
201
202## Authentication and operations
203
204Amazon Bedrock uses AWS-managed access controls. Your AWS administrator controls
205which accounts, roles, or temporary credentials can reach the supported model
206deployment. The exact authentication flow depends on the Bedrock configuration
207your organization uses.
208
209Plan for AWS-owned operational checks such as:
210
211- Account and model access configuration.
212- Region-specific deployment approval.
213- Temporary credential or token validity.
214- AWS quota, logging, and support workflows.
215
216## Pricing
217
218AWS bills Amazon Bedrock usage. Bedrock-specific pricing can differ from direct
219OpenAI API pricing, including regional processing premiums or other AWS-specific
220commercial terms.
221
222See [API pricing](https://developers.openai.com/api/docs/pricing) for direct OpenAI API pricing. For Bedrock
223pricing, use the AWS pricing materials published for the Bedrock deployment you
224plan to use.
225
226## Next steps
227
228- Confirm your supported model and AWS Region in Amazon Bedrock.
229- Verify the exact API features your workload needs.
230- Compare Bedrock pricing and direct API pricing before launch.
231- For Codex-specific setup, see
232 [Use Codex with Amazon Bedrock](https://developers.openai.com/codex/amazon-bedrock).