OpenAI AgentKit: Agents & Decisions API

OpenAI AgentKit helps developers build and ship AI agents; the Decisions API public beta adds typed answers on gpt-6-luna at $0.10 per 1M input tokens.

Update (Oct 17, 2026): Agent Builder shuts down Nov 30, 2026

OpenAI is deprecating Agent Builder, the drag-and-drop canvas in AgentKit. Per OpenAI’s deprecations page, developers were notified on June 3, 2026, and Agent Builder is scheduled to shut down on November 30, 2026. Existing users can keep using it until then. The notice covers Agent Builder only: ChatKit stays available, and OpenAI points developers to the Agents SDK.

What to use instead:

• Agents SDK (code): open your workflow, click Code, choose Agents SDK and copy the Python or TypeScript export into your app. OpenAI’s migration guide walks through it and warns that not every behaviour transfers unchanged.
• ChatGPT workspace agents (no code): on ChatGPT Business, Enterprise or Edu, paste the exported code into a new workspace agent and let the builder convert it.
• ChatKit: keep using it for the embedded chat front end.
• Other no-code builders: n8n, Zapier Agents and similar tools if you want a visual canvas. Our guide on how to build an AI agent compares the options step by step.

This OpenAI AgentKit is designed to address the common difficulties of fragmented toolchains and manual orchestration that slow down AI development. It targets AI engineers and developers across various industries, from tech startups to large enterprises like Klarna and Canva.

The platform’s primary value is its ability to reduce development time and manage the entire agent lifecycle, from initial prototype to production-level deployment. It offers an alternative to traditional, often difficult methods of building AI assistants by offering an integrated environment with visual builders, pre-built connectors, and specific evaluation tools, making the process faster and more efficient.

New: OpenAI Decisions API (public beta)

OpenAI’s Decisions API entered public beta for all developers. It returns typed answers — predicate probabilities, fixed choices or rubric scores — that OpenAI says arrive about 10x faster than the same model through the Responses API.

Use it when your app needs to classify, route or score an input, not when you need a long generated answer. The only model on Decisions today is gpt-6-luna via POST /v1/decisions.

OpenAI AgentKit Pricing (2026)

AgentKit itself is a toolkit around OpenAI’s platform APIs — you still pay for the models and endpoints you call. For Decisions specifically, OpenAI documents $0.10 per 1M input tokens on gpt-6-luna with no output, cache-read or cache-write charges on that endpoint (regional and long-context multipliers may apply). We verified this on the official Decisions guide on October 15, 2026.

Endpoint / product	Model	Price note
POST /v1/decisions	gpt-6-luna	$0.10 / 1M input only; no output charge
Responses / Chat (gpt-6-luna)	gpt-6-luna	Standard model pricing (input + output); not the Decisions-only rule
AgentKit builder tooling	—	Uses your OpenAI API usage; no separate AgentKit list price published on the pages we checked

Full how-to, question types and code examples: OpenAI Decisions API guide. Model context: GPT-6 Astra vs Sol. Safety: Are AI agents safe?.

Best Use Cases for OpenAI AgentKit

• Customer Support Automation: With this ai agent builder, support teams can construct agents to handle a high volume of customer inquiries. These agents can access knowledge bases to provide instant, factual answers, deflect common tickets, and escalate complex issues to human agents with full context. For example, an agent can handle two-thirds of all support conversations, as demonstrated by Klarna, freeing up human staff to focus on high-priority problems.
• Internal Operations & Knowledge Management: Companies can use OpenAI AgentKit to create internal assistants that help employees navigate complex processes and find information. An agent can connect to internal systems like SharePoint or Google Drive to answer questions about company policies, guide users through IT troubleshooting, or help analysts pull sales data. This centralizes knowledge and improves employee productivity.
• Sales & Marketing Automation: Sales teams can deploy OpenAI AgentKit to automate lead research and personalized outreach. An agent can analyze prospect data from various sources, generate tailored email drafts, and even score leads based on predefined criteria. This accelerates the sales cycle and allows representatives to focus on building relationships rather than on manual research.
• Data Analysis & Reporting: For business intelligence teams, this toolkit can build agents that automate data retrieval and report generation. A manager could ask an agent to “create a plan to improve sales in the Northwest region,” and the agent would analyze historical trends, seasonality, and external factors to produce a data-backed strategy, similar to the use case implemented by Albertsons.

AgentKit is still the OpenAI-shaped path to ship agents on their platform APIs. The Decisions API public beta is the fresh reason to revisit the page: typed routing answers on gpt-6-luna at $0.10 per 1M input with no output charge on that endpoint.

Use Decisions for classification and routing; keep Responses or Structured Outputs for generated objects and prose. Read our Decisions API guide before wiring production thresholds.