Accessing OpenAI fashions on Amazon Bedrock from Australia with world cross-Area inference

Australian groups working with OpenAI fashions can now entry the newest OpenAI fashions by way of Amazon Bedrock. Amazon Bedrock provides OpenAI GPT-5.6 Sol, Terra, and Luna with world cross-Area inference from each Asia Pacific (Sydney) and Asia Pacific (Melbourne) AWS Areas in Australia. Your utility calls the Amazon Bedrock Runtime endpoint in Asia Pacific (Sydney) or Asia Pacific (Melbourne), and Amazon Bedrock routes the request to a supported industrial AWS Area for processing. This gives entry to a broader capability pool with out requiring functions to handle vacation spot Area routing. GPT-5.6 Sol is suited to demanding reasoning, coding, and agentic workloads. Terra balances efficiency and value for on a regular basis manufacturing use. Luna gives quick, reasonably priced inference for high-volume and latency-sensitive functions. All three fashions settle for textual content and picture inputs, generate textual content, and assist context home windows of as much as 1 million tokens. With the Amazon Bedrock Runtime endpoint, you’ll be able to invoke these fashions with Responses API, Chat Completions API, and the Converse API from Asia Pacific (Sydney) and Asia Pacific (Melbourne).

On this put up, we present you find out how to use immediate caching to optimize your inference value, arrange Codex with OpenID Join (OIDC)-based authentication, and monitor utilization with Amazon CloudWatch and Coding Agent Insights.

GPT-5.6 world inference profiles

The next desk lists the three world profile IDs and the Australian supply Areas coated on this put up.

Profile membership and mannequin availability can change. Verify Cross-Area inference assist earlier than deployment.

Getting began

Earlier than continuing with this part, you’ll be able to comply with the GPT-5.6 weblog put up to arrange the next stipulations:

• An AWS account with Asia Pacific (Sydney) or Asia Pacific (Melbourne) enabled as your supply Area.
• In case your group makes use of service management coverage (SCP), confirm that the SCP permits the GPT-5.6 world inference profiles within the chosen supply Area.
• An AWS Identification and Entry Administration (IAM) position or consumer with correct permissions to invoke the GPT-5.6 inference profiles.
• Python 3.9 or later with the openai, boto3, and aws-bedrock-token-generator packages put in.

Within the following part, we present you find out how to confirm the lively world inference profiles with each the AWS Command Line Interface (AWS CLI) and the Amazon Bedrock console.

Utilizing the AWS CLI. The next instructions checklist the lively GPT-5.6 profiles and examine the Terra profile from the Sydney Area. You can even use the instructions with Sol or Luna by altering the inference profile ID. To run the identical examine from Melbourne Area, change ap-southeast-2 with ap-southeast-4.

aws bedrock list-inference-profiles 
--region ap-southeast-2 
--type-equals SYSTEM_DEFINED 
--query "inferenceProfileSummaries[?contains(inferenceProfileId, 'openai.gpt-5.6')].[inferenceProfileId,status]" 
--output desk

aws bedrock get-inference-profile 
--region ap-southeast-2 
--inference-profile-identifier world.openai.gpt-5.6-terra

Utilizing the Amazon Bedrock console. Open the Amazon Bedrock console, choose Sydney or Melbourne as your Area and select Inference profiles below Infer, and filter for World OpenAI GPT-5.6 Terra. The next screenshot reveals the lively profile from the Sydney supply Area.

Determine 1: The lively GPT-5.6 Terra world inference profile in Sydney

Invoke GPT-5.6 by way of Amazon Bedrock Runtime

GPT-5.6 helps three entry paths on the Amazon Bedrock Runtime endpoint: the OpenAI Responses API, OpenAI Chat Completions API, and Amazon Bedrock Converse API. The OpenAI-compatible APIs are known as on the /openai/v1 paths of this endpoint fairly than by way of the AWS SDKs. The endpoint accepts both AWS Signature Model 4 (SigV4) or an Amazon Bedrock mannequin inference API key. The next instance makes use of the AWS Bedrock Token Generator for Python to create a short-term Amazon Bedrock mannequin inference API key from the present AWS credentials, so the appliance doesn’t have to retailer a static key. You’ll be able to then create an OpenAI consumer utilizing this short-term API key to work with the supported APIs.

OpenAI Responses API. For functions that already use the OpenAI SDK with the Responses API, you’ll be able to level the consumer on the Regional Amazon Bedrock Runtime endpoint.

from aws_bedrock_token_generator import provide_token
from openai import OpenAI

area = "ap-southeast-2" # Use "ap-southeast-4" for Melbourne.
model_id = "world.openai.gpt-5.6-terra"
immediate = (
    "In three brief bullet factors, clarify how Availability Zones "
    "assist make an AWS utility extremely accessible."
)

openai_client = OpenAI(
    base_url=f"https://bedrock-runtime.{area}.amazonaws.com/openai/v1",
    api_key=provide_token(area=area),
)

responses_result = openai_client.responses.create(
    mannequin=model_id,
    enter=immediate,
    max_output_tokens=300,
)

print(responses_result.output_text)

For streaming output, set stream=True and iterate over the response occasions. The next instance prints textual content because it arrives:

response_stream = openai_client.responses.create(
    mannequin=model_id,
    enter=immediate,
    max_output_tokens=300,
    stream=True,
)

for occasion in response_stream:
    if occasion.kind == "response.output_text.delta":
        print(occasion.delta, finish="", flush=True)

OpenAI Chat Completions API. You can even work with the Chat Completions API in case your utility already makes use of that.

from aws_bedrock_token_generator import provide_token
from openai import OpenAI

area = "ap-southeast-2" # Use "ap-southeast-4" for Melbourne.
model_id = "world.openai.gpt-5.6-terra"
immediate = (
    "In three brief bullet factors, clarify how Availability Zones "
    "assist make an AWS utility extremely accessible."
)

openai_client = OpenAI(
    base_url=f"https://bedrock-runtime.{area}.amazonaws.com/openai/v1",
    api_key=provide_token(area=area),
)

chat_result = openai_client.chat.completions.create(
    mannequin=model_id,
    messages=[{"role": "user", "content": prompt}],
    max_completion_tokens=300,
    reasoning_effort="low",
)

print(chat_result.selections[0].message.content material)

Amazon Bedrock Converse API. Use Converse API when your utility calls Amazon Bedrock by way of an AWS SDK. Boto3 resolves credentials by way of the usual AWS credential chain.

import boto3

area = "ap-southeast-2" # Use "ap-southeast-4" for Melbourne.
model_id = "world.openai.gpt-5.6-terra"
immediate = (
    "In three brief bullet factors, clarify how Availability Zones "
    "assist make an AWS utility extremely accessible."
)
messages = [
    {
        "role": "user",
        "content": [{"text": prompt}],
    }
]

bedrock_client = boto3.consumer("bedrock-runtime", region_name=area)
converse_result = bedrock_client.converse(
    modelId=model_id,
    messages=messages,
    inferenceConfig={"maxTokens": 300},
)

print(converse_result["output"]["message"]["content"][0]["text"])

For streaming output, use converse_stream with the identical Area and profile ID, then iterate over the returned occasion stream.

stream_result = bedrock_client.converse_stream(
    modelId=model_id,
    messages=messages,
    inferenceConfig={"maxTokens": 300},
)

for occasion in stream_result["stream"]:
    if "contentBlockDelta" in occasion:
        delta = occasion["contentBlockDelta"]["delta"]
        if "textual content" in delta:
            print(delta["text"], finish="", flush=True)
print()

The previous examples name Amazon Bedrock in Asia Pacific (Sydney). To run the identical instance for Asia Pacific (Melbourne), set the Area to ap-southeast-4 and run the code.

Working with immediate caching

GPT-5.6 immediate caching is out there by way of the supported APIs. GPT-5.6 helps two caching modes on Amazon Bedrock. Implicit caching is enabled by default, and no code adjustments are required, whereas with express caching, you’ll be able to outline the reusable prefix, cache boundary, and cache key. The GPT-5.6 weblog put up gives examples that illustrate the immediate caching functionality.

Organising Codex with GPT-5.6 on Amazon Bedrock

Codex can use the identical world inference profiles by way of Amazon Bedrock Runtime. Set up the newest Codex CLI to make use of the native Amazon Bedrock Runtime mannequin supplier. Right here, we validated the next configuration with codex-cli 0.149.1 utilizing GPT-5.6 Sol from Asia Pacific (Sydney).

npm set up -g @openai/codex@alpha
codex --version

For organizations whose id supplier is Okta, Auth0, Microsoft Entra ID, Amazon Cognito, or AWS IAM Identification Middle, the AWS OIDC Auth Helper repository gives a pattern credential helper. First, comply with the information to configure your id supplier, the corresponding AWS federation useful resource, and an IAM position carrying the Amazon Bedrock permissions proven earlier. Then add a named profile to ~/.aws/config, so the helper doesn’t change credentials resolved by the default profile. Change the placeholders with the trail to the put in helper and the profile title outlined within the ~/.aws/config file.

[profile ]
credential_process =  --profile 
area = ap-southeast-2
output = json

This federation helper exchanges an OIDC token for non permanent AWS credentials, and Codex reads by way of the usual AWS credential chain with no additional configuration. Subsequent, create or replace ~/.codex/config.toml and reference the AWS profile, see the Codex configuration reference for different supported settings:

mannequin = "world.openai.gpt-5.6-sol"
model_provider = "amazon-bedrock-runtime"
model_reasoning_effort = "excessive"

[model_providers.amazon-bedrock-runtime.aws]
profile = ""
area = "ap-southeast-2"

If the helper doesn’t have a legitimate cached session, it opens the configured sign-in web page in your browser. After you authenticate, the helper returns non permanent AWS credentials by way of credential_process. Requests are signed with AWS SigV4, so no API secret is concerned within the inference path. When the profile is backed by AWS IAM Identification Middle, the credentials are already short-term and rotate with the only sign-on session. To make use of this with Asia Pacific (Melbourne) Area, set the Area to ap-southeast-4 within the AWS profile and Codex configuration.

Quota administration

GPT-5.6 on-demand quotas are measured in requests per minute (RPM) and tokens per minute (TPM). Token burndown determines how every request consumes TPM. Token consumption is calculated from enter tokens, cache-write enter tokens, and output tokens multiplied by the mannequin’s burndown charge. As an example, for GPT-5.6, enter tokens and cache-write enter tokens depend at 1:1, whereas every output token consumes 10 tokens from the quota. Evaluate the GPT-5.6 quotas within the Service Quotas console from the supply Area your utility makes use of: Asia Pacific (Sydney) Area (ap-southeast-2) or Asia Pacific (Melbourne) Area (ap-southeast-4). Earlier than manufacturing rollout, request will increase early, monitor quota utilization, and take a look at consultant prompts, output lengths, streaming conduct, concurrency, and peak visitors. See Amazon Bedrock quotas for the present values and token burndown charges.

Monitoring and logging

As a result of GPT-5.6 requests use the Amazon Bedrock Runtime API, requests made by way of the worldwide inference profiles seem in mannequin invocation logging like different on-demand requests. When logging is enabled, information embody the mannequin or inference profile ID used for the decision and invocation metadata. Codex makes use of OpenTelemetry (OTel) and exports metrics over OTLP/HTTP, see arrange OpenTelemetry for OpenAI Codex for extra particulars.

CloudWatch Coding Agent Insights gives a dashboard for Codex telemetry, together with token utilization, API requests, lively customers, dialog exercise, and elective organizational dimensions. There are two paths to configure Coding Agent Insights in CloudWatch for Codex, that’s, utilizing Bearer token or Enterprise rollout.

The next instance reveals find out how to configure the Coding Agent Insights for Codex utilizing the Bearer method. First, get a CloudWatch metrics API key, then add the next sections to ~/.codex/config.toml.

[otel]
surroundings = "manufacturing"

[otel.metrics_exporter]
otlp-http = { endpoint = "https://monitoring.ap-southeast-2.amazonaws.com/v1/metrics", protocol = "binary", headers = { "Authorization" = "Bearer YOUR_CLOUDWATCH_METRICS_API_KEY" } }

Change YOUR_CLOUDWATCH_METRICS_API_KEY with the important thing created in CloudWatch, then begin Codex. This CloudWatch metrics API key can then authorize the export to the Regional CloudWatch endpoint. After telemetry arrives, open the CloudWatch console in Asia Pacific (Sydney) Area, select GenAI Observability, Coding Agent Insights, and the Codex tab. You will note the Coding Agent Insights dashboard shows Codex token utilization, request exercise, cache hit charge, and extra.

Determine 2: Codex token and request exercise in CloudWatch Coding Agent Insights in Asia Pacific (Sydney)

To populate the Group, Setting, Division, Price Middle, Location, Staff, and Consumer filters, present the corresponding values by way of OTEL_RESOURCE_ATTRIBUTES. AWS classifies that CloudWatch metric API key as a long-term credential and recommends it solely the place short-term AWS credentials will not be possible. Deal with config.toml as a secret and prohibit its file permissions.

For organizations that federate developer id by way of company single sign-on, we advocate utilizing the enterprise rollout, the place an area collector indicators the export with SigV4 utilizing the developer’s federated credentials and no token is distributed.

Conclusion

On this put up, we confirmed find out how to uncover and invoke the GPT-5.6 Sol, Terra, and Luna world inference profiles from Asia Pacific (Sydney) and Asia Pacific (Melbourne) Areas. We additionally launched find out how to configure Codex to make use of Amazon Bedrock Runtime and export Codex telemetry to CloudWatch Coding Agent Insights.

To get began, comply with the examples on this put up and take a look at GPT-5.6 fashions in your account. In case you are utilizing Codex, you’ll be able to configure the Amazon Bedrock Runtime supplier and allow CloudWatch Coding Agent Insights within the AWS Console to watch your Codex consumption. For pricing particulars, see Amazon Bedrock pricing.

Concerning the authors