AI・テクノロジー
OpenAI's "Operator API" Goes Live — Web Automation Delivered via API Reshapes the RPA Market
機械翻訳 / Machine-translated
On August 26, 2026, OpenAI officially launched the "Operator API," which allows external systems to call browser operations, form filling, and data extraction capabilities. Rather than a flat monthly fee, the service adopts a pay-per-task pricing model, and its structure directly targets the routine task automation space that UiPath and Automation Anywhere have long dominated.
According to the official documentation, the Operator API offers three endpoints: "browser_action" (browser operations), "form_fill" (form filling), and "data_extract" (data extraction), with a GPT-4o-class visual reasoning model running on the backend.
Pricing is usage-based per task execution — $0.12 for browser_action and $0.08 for data_extract. Integration with AWS Marketplace and Azure AI Studio was confirmed on the day of the official launch.
"All we had to do was define the business workflow in JSON, and Operator handled the actual browser operations for us. We no longer need the dedicated engineer that RPA implementation used to require." (DX Manager at a mid-sized manufacturer, via X)
OpenAI revealed that during the nine-month beta period beginning November 2025, 72% of the 340 participating companies indicated they were "considering a full migration away from their existing RPA solutions."
OpenAI first introduced Operator to ChatGPT Plus users in January 2025. Initially it remained little more than a demo feature, but the enterprise beta gradually expanded, leading to the general API launch in August 2026.
The push toward an API offering was driven by enterprise demand to "embed the capability within their own systems." Requests surged from companies wanting to integrate Operator functionality directly into their internal business workflows — rather than through the ChatGPT UI — revealing a level of demand that could not be met through a standalone application alone.
The global RPA market was valued at approximately $3.8 billion in 2025, with projected annual growth of 22%. However, several consulting firms had forecast a "structural shift between 2027 and 2028 driven by the rise of AI agents," and the Operator API is positioned as a move that accelerates that transition.
The leading enterprise RPA platform, UiPath Enterprise, is primarily sold under fixed licensing starting at $28,000 per year. Because the Operator API charges based on the number of tasks executed, it offers strong economic appeal for SMEs with irregular workloads and for large enterprises considering partial adoption for specific processes. The day after the announcement, UiPath's stock fell 7.3%.
Conventional RPA depends on screen coordinates and CSS selectors, requiring script updates every time a UI changes. The Operator API claims resilience to UI changes by combining GPT-4o's image recognition with intent estimation — though the actual reduction in maintenance costs will need to be validated through accumulated real-world operational data.
Risk of violating the Terms of Service (ToS) of web services accessed by the Operator API is managed at each company's own responsibility. Going live in heavily regulated sectors such as finance and healthcare is expected to require additional design work, including data handling policies and audit trails for operation logs.
The official launch of the Operator API accelerates the transition from "AI agent = a specific LLM application" to "AI agent = a component of business infrastructure."
Looking at the pricing structure, $0.12 per task will, in many current cases, be more expensive than both labor costs and RPA SaaS. However, referencing OpenAI's track record on API price revisions — the input price for GPT-4 has fallen more than 90% compared to 2023 — a scenario in which the per-task price halves by 2027 is entirely realistic. If that happens, the breakeven point shifts considerably.
Competitors are not standing still. Anthropic already offers a Computer Use API, and multiple developers have reported signs that Google is moving toward an API release for Project Mariner. If all three major players have computer-operation APIs on the market by the end of 2026, full-scale price competition is likely to begin in the latter half of that year.
Whether enterprises should commit to a full migration right now depends on the regulatory complexity of their workflows and the payback status of their existing RPA investments. That said, the cost of "trying it out on just one process" is now clearly lower than before. The realistic interpretation is that the threshold for running a pilot has dropped.
The Operator API has become the first major example of "AI that operates UIs" being extracted and offered as business infrastructure. It will work to accelerate the restructuring of the RPA market and shift the cost structure of routine task automation.
The next focal point is when price competition between the two rivals — Anthropic's Computer Use API and Google's Mariner API — will truly heat up. The window in which OpenAI can hold its first-mover advantage is estimated to last until the end of 2026.
This article was written by an AI writer (AI News) from the Mirai News editorial team.
