Quickly Get Started with OpenAI GPT‑5.5, GPT‑5.4, and Codex on Amazon Bedrock

Overview

OpenAI GPT‑5.5, GPT‑5.4 and Codex are now available on Amazon Bedrock. The models can be invoked via the Bedrock Responses API or through the OpenAI programming Agent.

Codex programming agent

Codex is an AI‑driven software‑development agent. OpenAI reports that more than 4 million developers use Codex weekly for code writing, refactoring, debugging, testing and verification in large codebases. Switching the inference base to GPT‑5.5 raises Codex’s capability for complex, long‑running workflows. Access methods include the Codex App, Codex CLI, and IDE plugins for VS Code, JetBrains and Xcode; all inference requests are routed through the Bedrock Responses API.

Authentication

Set environment variables for the OpenAI SDK: OPENAI_BASE_URL, OPENAI_API_KEY (your Bedrock API key), and BEDROCK_OPENAI_MODEL_ID (e.g., openai.gpt-5.5).

Bedrock supports two credential modes: an explicit Bedrock API key supplied via AWS_BEARER_TOKEN_BEDROCK or the default AWS SDK credential chain.

Configuration

Edit ~/.codex/config.toml (or ~/.codex/.env) to set the preferred AWS region and model ID ( openai.gpt-5.5, openai.gpt-5.4, openai.gpt-oss-120b, openai.gpt-oss-20b). Restart the desktop app or VS Code extension after changes. The CLI shows status on the /status page.

Python example (OpenAI SDK)

import os
from openai import OpenAI

client = OpenAI(
    base_url=os.environ["OPENAI_BASE_URL"],
    api_key=os.environ["OPENAI_API_KEY"],
)

response = client.responses.create(
    model=os.environ["BEDROCK_OPENAI_MODEL_ID"],
    input=[
        {"role": "developer", "content": "You are a software engineer with excellent AWS cloud knowledge. Be concise and practical."},
        {"role": "user", "content": "Design a distributed architecture on AWS in Python that should support 100k requests per second across multiple geographic regions."},
    ],
    reasoning={"effort": "medium"},
    text={"verbosity": "low"},
)
print(response.output_text)

Curl example

curl "$OPENAI_BASE_URL/responses" \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $OPENAI_API_KEY" \
  -d '{
    "model": "openai.gpt-5.5",
    "input": [
      {"role": "developer", "content": "You are a software engineer with excellent AWS cloud knowledge."},
      {"role": "user", "content": "Design a distributed architecture on AWS in Python that should support 100k requests per second across multiple geographic regions."}
    ],
    "reasoning": {"effort": "medium"},
    "text": {"verbosity": "low"}
  }'

Responses API capabilities

Supports multi‑turn conversation state, hosted tools, function calling, complex tool orchestration, and long‑running background tasks. Additional examples are in the OpenAI Cookbook.

Model latency and effort settings

OpenAI positions GPT‑5.5 as the fastest model and GPT‑5.4 as a balanced speed‑cost option. Observed latency depends on inference intensity, output length, tool calls, background mode, region, quota, throttling, prompt size and cache hit rate. Recommended: start GPT‑5.5 with effort":"medium"; for GPT‑5.4 explicitly set the effort parameter instead of relying on its default none.

Elastic scaling on Bedrock

The new Bedrock inference engine is built for rapid cross‑model scaling. Steady‑state workloads are maintained while capacity is increased on demand. Requests that exceed current capacity are queued rather than rejected.

Regional availability

GPT‑5.5: US East (Ohio).

GPT‑5.4: US East (Ohio) and US West (Oregon).

Additional regions will be added over time.