OpenAI has opened its Agents API to all developers, combining durable sessions, tool use, optional subagents and a choice of hosted or external execution environments.
RELATED BUYER PROFILEReview the Agents API access, cost and test-pending profilePricing · access · strengths · limitations →CONTINUE THE DECISIONDefine the agent permission and approval boundaryEvidence · workflow · next action →CONTINUE THE DECISIONPlan a bounded same-task agent pilotEvidence · workflow · next action →CONTINUE THE DECISIONCalculate complete cost per accepted taskEvidence · workflow · next action →What you need to know
- The Agents API is in public beta and available to all developers
- OpenAI charges no separate Agents API fee; customers pay for the models, tools and compute their agents consume
- Developers can use an OpenAI-hosted sandbox, their own infrastructure or an integrated sandbox provider
The agent operating model
What OpenAI launched
The Agents API exposes the managed harness and infrastructure used by Codex through a developer API. A session can specify the task, model, tools, vaults and execution environment. OpenAI says the service is designed for work that can span hours or days, with context management, intermediate outputs and recovery handled as part of the agent runtime.
The environment choice changes the risk boundary
Teams can run an agent inside an OpenAI-hosted sandbox, on their own infrastructure or through an integrated sandbox provider. Hosted convenience does not remove the need to define filesystem, network, secret and tool permissions. Self-hosting may improve control for some workloads, but it also moves isolation, patching, observability and capacity responsibility back to the buyer.
Tool search, programmatic calls and subagents
The harness can load relevant tool definitions on demand, run supported calls in parallel and delegate independent work to subagents. These features can reduce hand-built orchestration, but they can also multiply calls, compute and failure paths. Production evaluation should record every model, tool, environment and subagent invocation behind a completed task.
Price is usage, not a free agent claim
OpenAI says there is no additional fee for the Agents API. Customers pay for the tokens and tools their agents use, while sandbox or partner infrastructure can introduce further costs. A useful comparison therefore measures complete-task cost, including retries, subagents, tool calls, environment runtime, storage, review and failed outcomes—not only the selected model's token rate.
HubAI buyer verdict
Agents API is a serious shortlist candidate for teams that want the Codex-style harness without maintaining their own durable orchestration layer. Public beta status, evolving behaviour and vendor dependence make a bounded pilot essential. Freeze one multi-step task, permissions and acceptance test; compare it with the current agent stack on success rate, recovery, human intervention, latency and total cost before migration.
HUBAI VIEWThe buying decision is whether the managed harness removes enough orchestration work to justify provider dependence, tool costs and the security review required for long-running agents.
Buyer decision signal: Public beta · managed cloud agents
What to verify next
1Choose the execution environment and document who owns its security controls
2Restrict filesystem, network, secrets and tool access to the smallest required set
3Record every model, tool, subagent and sandbox cost per completed task
4Test cancellation, timeout, recovery and partial-output behaviour
5Verify data retention, regional availability and contractual controls for the intended workload
6Compare the managed harness with the existing orchestration stack on the same acceptance test
7Keep production rollout gated while the service remains in public beta
Read the evidence
Capabilities, availability and prices can change. HubAI keeps analysis separate from the underlying official material.
01OpenAI: Introducing the Agents APIOpen source ↗02OpenAI Developers: Agents API overviewOpen source ↗03OpenAI API pricingOpen source ↗
Products
Products
Code AI
Image AI