OpenAI review: Agents API, Agents SDK and Frontier as an enterprise agent platform

Answer in brief: OpenAI offers three layers for enterprise agents. The Agents API is a public beta open to all developers: OpenAI runs a managed Codex harness with durable sessions, sandboxes, MCP connections and subagents. The Agents SDK runs the agent loop inside your own application. Frontier is OpenAI's sales-led enterprise platform for deploying and governing agents across a company. On our market map OpenAI is an independent, model-native platform. The Agents API has no separate fee: you pay model token rates (for example, gpt-6.1-sol at $2 input / $10 output per 1M tokens) plus tool and sandbox charges. Frontier has limited availability and no published price.

Top pick for teams building on OpenAI models. A managed harness that handles orchestration, context compaction and recovery, plus sandboxes that run code and produce files. Frontier adds per-agent identity, permissions and OpenAI Forward Deployed Engineers. Visit OpenAI →

Independent research, last checked November 9, 2026. We have not tested these products hands-on; facts come from OpenAI's own pages and documentation.

Where it sits on the market map

OpenAI is the model maker turned agent platform. Its advantage is close integration with its own models. Its trade-off is that the stack centers on them, while Google and AWS sell model choice.

What it does for enterprise agents

Agents API (beta): sessions, orchestration, context compaction and recovery are managed by OpenAI. Agents can run commands in OpenAI-hosted or self-hosted sandboxes, connect to MCP servers, delegate to subagents and resume where they left off.
Agents SDK: your application controls deployment, storage, approvals and runtime integration, and the SDK runner handles the agent loop and handoffs.
Frontier: a shared business-context layer over warehouses, CRM, ticketing and internal apps. Agents execute with memory across local, enterprise-cloud and OpenAI-hosted environments. It adds built-in evaluation, a distinct identity and explicit permissions per agent, and guardrails.
ChatKit: an embeddable chat experience that remains available.

Who it fits

Product and engineering teams that want OpenAI's models with minimal agent plumbing, and large enterprises that want to work directly with OpenAI on company-wide agents.

Strengths that win deals

Low integration effort with the Agents API, according to OpenAI's own comparison.
No platform fee on the Agents API, so you pay only for usage.
Frontier works with existing systems across multiple clouds and does not require replacing agents you already deployed.
An ecosystem of named Frontier partners and alliances, including McKinsey, BCG, Accenture and Capgemini.

Honest drawbacks

The platform is changing fast. Agent Builder and the Evals platform shut down on November 30, 2026, and existing Evals became read-only on October 31, 2026. OpenAI points users to the Agents SDK, ChatGPT Workspace Agents and Promptfoo.
The Agents API is still in beta.
Frontier has limited, sales-led availability and no public price.
The stack is built around OpenAI's own models.

Pricing model

Agents API: the selected model's API rates, plus OpenAI tools at standard rates and hosted sandboxes at container rates.
Example model rate: gpt-6.1-sol, $2 input / $10 output per 1M tokens (standard, short context).
Uplifts: regional data-residency endpoints carry a 10% uplift for models released on or after March 5, 2026.
Frontier: contact OpenAI sales.

Customers OpenAI names

OpenAI named HP, Intuit, Oracle, State Farm, Thermo Fisher and Uber as initial Frontier adopters. It said BBVA, Cisco and T-Mobile had piloted Frontier's approach.

How to evaluate OpenAI's stack

Pick your runtime: the Agents API (OpenAI-managed), the SDK (you manage) or the Responses API (build from scratch).
If you used Agent Builder, export workflows and plan the migration before November 30, 2026.
Model token spend at your expected volume, including cached input.

Best alternatives

Amazon Bedrock AgentCore: runs the OpenAI Agents SDK, with any model.
Google Gemini Enterprise Agent Platform: a managed runtime with 200+ models.
Microsoft Copilot Studio: low-code agents with Foundry model access.

FAQ

Is the OpenAI Agents API generally available?

No. OpenAI calls it a public beta available to all developers and says it is working toward general availability.

Is OpenAI Frontier generally available?

OpenAI launched Frontier with limited availability and sells it through sales; we found no official announcement of general availability as of November 9, 2026.

What replaces Agent Builder?

OpenAI recommends the Agents SDK for code-built agents and ChatGPT Workspace Agents for natural-language building. ChatKit remains available.

Does the Agents API charge a platform fee?

No. You pay model token rates plus tool and hosted-sandbox charges.

Sources

Visit OpenAI → · Visit OpenAI Frontier →

← Back to the ranked list of enterprise agent platforms · Pricing comparison