OpenAI GPT-5.5, GPT-5.4, and Codex are now generally available on Amazon Bedrock, AWS confirmed on June 1, 2026. Six weeks after the limited preview launched on April 28, any AWS account can now route OpenAI inference through Bedrock at the same per-token pricing OpenAI charges directly, with usage drawing down existing AWS cloud commitments.

Try It: Swap OpenAI Direct for Bedrock In One Config Line

If you already build on AWS, here is the 15-minute migration that unlocks the discount most teams will actually feel:

  1. Open the Bedrock OpenAI model catalog in your AWS console and request access to gpt-5.5, gpt-5.4, and codex-1.
  2. Replace the OpenAI base URL in your SDK config with the Bedrock Runtime endpoint for your region. Model IDs follow the openai.gpt-5-5-v1:0 pattern.
  3. Update IAM to grant bedrock:InvokeModel on the new ARNs. Reuse the same key your team already audits through CloudTrail.
  4. For Codex specifically, the same key works inside the Codex App, the Codex CLI, and the official Codex VS Code extension. No second login.

The payoff is not raw price (Bedrock matches OpenAI list rates), it is procurement: token spend now counts against the same AWS commit your S3 and EC2 invoices already consume.

Why It Matters

For two years, picking GPT for a production agent meant a second vendor relationship outside the hyperscaler the rest of the stack lived on. Azure OpenAI Service solved that for Microsoft shops; AWS customers had to wire OpenAI as a third-party SaaS and explain it on their next compliance review. That just ended.

The June 1 GA also brings Managed Agents powered by OpenAI out of preview, so the agentic orchestration layer (tools, memory, guardrails) is fully managed by AWS instead of rebuilt on top of raw chat completions.

Key Details

  • Models GA: GPT-5.5, GPT-5.4, and Codex. Smaller GPT variants and o-series reasoning models stay in preview for now.
  • Pricing: Matches OpenAI first-party rates with no Bedrock surcharge, per Amazon's June 1 statement. Usage applies to existing AWS commit dollars.
  • Codex surfaces: Codex App, CLI, and VS Code extension all route inference through Bedrock when configured against an AWS account.
  • Security: IAM, KMS encryption, VPC endpoints, and CloudTrail logging on every call, same as native Bedrock models.
  • Multi-model: Sits beside Anthropic, Meta, Mistral, Cohere, and Amazon Nova in the same Bedrock console, so swapping between providers is a model-ID change rather than a vendor migration.

What to Do Next

If your team is on AWS and currently splits inference between Bedrock and a direct OpenAI key, consolidate. The commit-drawdown alone usually pays for the migration sprint. If you build agents that call GPT-5.5 and Claude in the same workflow, the Bedrock console is now the single pane for both, with one IAM policy instead of two API keys to rotate. The Codex VS Code extension is the lowest-friction onboarding path: install it, point it at your Bedrock-enabled AWS profile, and the same agent runs against your AWS budget.