# OpenAI's Agents API Turns the Codex Harness Into a Managed Service, With No Platform Fee

OpenAI opened the Agents API in public beta on 10 September: the Codex harness as a managed service, with no fee beyond tokens and tools. No platform fee is not the same as cheap.

By Neomanex. Updated 2026-09-23.

**Our take: prototype on it, keep production off it.** If you run the Codex harness on your own infrastructure for cloud agents, price one real workload on the [OpenAI Agents API](/directory/openai-agents-api) now, because it removes the hosting work at no platform fee. Do not move production agents yet: it is a public beta. That matches its **conditional** rating in our directory. [OpenAI Codex](/directory/openai-codex), the product built on the same harness, stays **recommended**.

## What happened

OpenAI opened the Agents API to every developer on 10 September 2026, per its [API changelog](https://platform.openai.com/docs/changelog). From the [announcement](https://openai.com/index/introducing-the-agents-api):

- **Access.** "Agents API is available in public beta today to all developers."
- **Engine.** "The Agents API is powered by the open-source Codex harness."
- **Pricing.** "There are no additional fees for using the Agents API – you simply pay for the tokens and tools your agents use."
- **Where it runs.** "You choose the agent's compute environment: in an OpenAI-managed sandbox, on your own infrastructure, or with one of our sandbox partners." The page also lists "Fully managed environments or deployments within your VPC".
- **Built in.** Context management with automatic compaction across long sessions, tool search to cut token usage, programmatic tool calling with parallel execution, and multi-agent support with subagents. Tools include MCP, custom functions and web search.
- **Sandbox partners.** Blaxel, Cloudflare, Daytona, DigitalOcean, E2B, Modal, Oracle, Runloop and Vercel.

OpenAI says it will "iterate quickly" on feedback during the beta, so expect the surface to move.

## Why it matters

Codex is the finished product: hand it a task, get a change back. The Agents API exposes the harness underneath, so you build your own agents on it instead of using OpenAI's, and you stop maintaining the orchestration layer yourself. OpenAI pitches the subagent support as parallel work "without requiring you to build your own orchestration."

No platform fee is not the same as cheap. The bill is entirely tokens and tool calls, and subagents plus long-running sessions are exactly the features that multiply token spend. Automatic compaction and tool search claw some of that back; whether compaction costs accuracy on your workload is something you will need to measure. The announcement gives no per-session cost guidance. We have not run it.

## What changes for you

| If you... | Do this |
|---|---|
| Self-host the Codex harness | Run one representative workload on the Agents API and compare the token bill with your infrastructure cost |
| Plan to use subagents | Set per-session token budgets first: spend scales directly with agent activity |
| Already use a sandbox provider | Check it is on the partner list before migrating |
| Run production agents today | Leave them where they are until the API leaves public beta |

## FAQ

**Does the Agents API cost extra?** No platform fee. You pay for the tokens and tools your agents use.

**Is it production-ready?** It is a public beta, open to all developers. We rate it conditional: fine for prototypes, not yet for production.

**Can it run inside my own infrastructure?** Yes. OpenAI offers an OpenAI-managed sandbox, your own infrastructure, or a sandbox partner.

Canonical: https://neomanex.com/news/openai-agents-api-public-beta-codex-harness