<!-- mobian-agent-page publisher="dailydev" canonical="https://daily.dev/posts/96Gc6DVzO" -->

---
title: OpenAI's Agents API adds computer use with a hosted browser
description: OpenAI's Agents API now supports computer use, letting developers spin up a browser and an agent together through a single API call instead of wiring up...
canonical: https://daily.dev/posts/openai-s-agents-api-adds-computer-use-with-a-hosted-browser-96gc6dvzo
twitter:card: summary_large_image
twitter:site: @dailydotdev
og:type: website
og:site_name: daily.dev
og:title: OpenAI's Agents API adds computer use with a hosted browser | daily.dev
og:description: OpenAI's Agents API now supports computer use, letting developers spin up a browser and an agent together through a single API call instead of wiring up...
og:url: https://daily.dev/posts/openai-s-agents-api-adds-computer-use-with-a-hosted-browser-96gc6dvzo
og:image: https://api.daily.dev/og/posts/96Gc6DVzO.png
og:image:alt: OpenAI's Agents API adds computer use with a hosted browser
og:image:width: 1200
og:image:height: 630
og:locale: en
---

> ## Documentation Index
> Fetch the complete documentation index at: https://daily.dev/llms.txt
> Use this file to discover all available pages before exploring further.

# OpenAI's Agents API adds computer use with a hosted browser

**[Collections](https://daily.dev/sources/collections)** · 2 min read · 1 upvotes · 1 comments

## Summary

OpenAI's Agents API now supports computer use, letting developers spin up a browser and an agent together through a single API call instead of wiring up browser automation separately. The feature was announced via a social thread rather than detailed documentation, and reaction from AI practitioners has focused on the convenience of bundling rather than novelty, since similar computer-use capabilities already exist elsewhere. Exact behavior of the hosted browser remains unclear pending further documentation.

## Content

Pine AI has launched Pine Computer, a cloud computer, harness, and runtime layer built for agentic tasks. Access is through a private beta waitlist from @PineAIAssistant.

## How it works

Pine Computer is a computer-in-the-cloud-as-a-service. As @svpino describes it, your application gets a cloud computer with a built-in AI agent. The flow looks like this:

- Your app uses Pine's SDK to create the cloud computer.
- You give the built-in agent a task in plain English.
- The agent has access to a browser, files, and a shell.
- Your app gets notifications while the agent works.
- You get the outputs back when it's done.

The agent plans and executes the work across websites, files, and apps, instead of driving a computer designed for humans. Pine Computer also doesn't simulate clicks and screenshots. It reads web pages as structured data and reports changes directly to the AI. @svpino notes it looks like Computer Use from existing agents, but says the way these capabilities are packaged is really cool.

## Benchmark results

The numbers come from SaaS-Bench v1.1, which covers 106 multi-step tasks inside 23 business apps. The checkpoint score counts the steps passed along the way.

- Pine Computer has the highest checkpoint score in the publisher's table: 78.3%, against 74.3% for Opus 5 with Claude Code.
- It completed fewer whole tasks: 27.4% against 31.1%.
- Running GPT-5.6 Luna, Pine reports about 1/20 the model-token cost of GPT-5.6 Sol plus Codex.
- It reports 1/26 the model-token cost of Opus 5 plus Claude Code.
- In selected preliminary internal tests, it was 2 to 5 times faster, depending on the task.

These are Pine's own published figures. The split between checkpoint score and whole-task completion is worth keeping in mind: Pine leads on partial progress but trails on finishing entire tasks.

## Why it matters

@rohanpaul_ai calls it another release showing that models need better harnesses and environments. I think that's the right read. The interesting part isn't a new model, it's the environment built around one, and the cost numbers suggest a lot of the gain comes from there.

OpenAIDevs also shared a note from @stevendcoffey about a computer use feature in the Agents API that spins up a browser plus an agent, so this kind of packaging is showing up in more than one place.

## Questions this post answers

### What does the computer use feature in OpenAI's Agents API actually do?

It lets developers spin up a browser and an agent together through a single API call, rather than manually wiring up browser automation for an agent separately. The capability to click through a browser isn't new for agents generally, but bundling the browser and agent into one hosted call changes the setup work required. OpenAI has not published detailed documentation on how the hosted browser behaves in practice.

_Developers tracking agent tooling changes like this follow practical breakdowns on daily.dev._

## Community take

How the wider developer community reacted, aggregated from 4 discussions and 20 comments across x (as of 2026-10-10).

**TL;DR:** Reactions focus on the gap between Pine's high checkpoint score and its lower whole-task completion rate, with many arguing that finishing jobs end-to-end matters more than partial progress, while others see the agent-harness approach (structured data instead of screenshots) as a promising direction.

**Sentiment:** 25% positive · 45% mixed · 30% skeptical

**The case for**

- Reading pages as structured data instead of screenshots is seen as a good way to target the stalling-on-long-tasks failure mode.
- Persisting state across sessions instead of losing context on tab close is viewed as a real improvement over typical agent SDKs.
- Some find the low token cost and speed claims impressive and see the harness, not the base model, as where gains are coming from now.

**The pushback**

- Many note the gap between a high checkpoint rate and a much lower completed-task rate means runs often stall before finishing.
- Some argue checkpoint-based partial credit is misleading and prefer cost or success measured per completed task rather than per token.
- Security and access concerns are raised about giving agents full cloud/browser access to logins and accounts.
- Skepticism that the benchmark gap (78.3% vs 74.3%) is meaningful without independent verification, since the only numbers available are from the publisher itself.

**By community**

- x (mixed): Discussion is split between excitement over the harness-over-model framing and skepticism about whether high checkpoint scores with low task completion actually represent progress.

**Hottest debate:** Whether a high checkpoint-completion score is meaningful progress when the whole-task completion rate is lower than a competitor's.

**Open questions**

- Has anyone independently verified Pine's SaaS-Bench v1.1 numbers?
- How does the agent handle site logins and account security when given full cloud/browser access?
- What is the cost per successfully completed task, not just per token?

**Highlights**

> @testingcatalog the gap between 78% checkpoints and 27% completions is the real signal - the harness makes early progress then stalls on the long tail. reading pages as structured data instead of screenshots should target exactly that failure mode.
> — [siftloom on x](https://x.com/siftloom/status/2108621139824050499)

> @testingcatalog checkpoint score and finished tasks are telling different stories here. if you trial it make the agent write a done check for the last step not just the steps it passed. partial credit looks great until the job is unfinished
> — [aishortcut42 on x](https://x.com/aishortcut42/status/2108609485476622506)

> @testingcatalog 78.3% vs 74.3% from the publisher's own table is a thin gap to call it. has anyone run SaaS-Bench v1.1 independently yet, or is pine's number the only one out there?
> — [jdn1el on x](https://x.com/jdn1el/status/2108599400083706111)

> @rohanpaul_ai 1/20 the token cost is impressive. The more important metric is cost per successfully completed task. Lower token costs mean little if the agent fails more often.
> — [48brief on x](https://x.com/48brief/status/2108760348769796500)

> @rohanpaul_ai ai agents with full cloud access sounds like a great way to accidentally empty my bank account.
> — [chenzhangreal on x](https://x.com/chenzhangreal/status/2108678363649016282)

**Source threads**

- [x](https://x.com/testingcatalog/status/2108595450529857856) · 1 points · 14 comments
- [x](https://x.com/rohanpaul_ai/status/2108677451471794275) · 0 points · 6 comments
- [x](https://x.com/svpino/status/2108774641284341979) · 0 points · 0 comments
- [x](https://x.com/rohanpaul_ai/status/2108752533569114485) · 0 points · 0 comments

## Community discussion

Top comments from developers on daily.dev.

**@yosinhonbotirov** · 0 upvotes

> oh that's good

## Similar posts on daily.dev

- [OpenAI launches dots, always-on AI agents with their own cloud computers](https://daily.dev/posts/openai-launches-dots-always-on-ai-agents-with-their-own-cloud-computers-ci9onynya) · The Next Web · 0 upvotes · 0 comments
- [Your AI Agent Now Acts Without Being Asked. Here’s What That Changes for Builders](https://daily.dev/posts/your-ai-agent-now-acts-without-being-asked-here-s-what-that-changes-for-builders-k39ulcl6g) · Medium · 0 upvotes · 1 comments

---

Tags: [#ai-agents](https://daily.dev/tags/ai-agents), [#openai](https://daily.dev/tags/openai)

[View this post on daily.dev](https://daily.dev/posts/openai-s-agents-api-adds-computer-use-with-a-hosted-browser-96gc6dvzo)

```json
{"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]}
{"@context":"https://schema.org","@type":"TechArticle","headline":"OpenAI's Agents API adds computer use with a hosted browser","url":"https://daily.dev/posts/openai-s-agents-api-adds-computer-use-with-a-hosted-browser-96gc6dvzo","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/openai-s-agents-api-adds-computer-use-with-a-hosted-browser-96gc6dvzo"},"datePublished":"2026-10-07T05:11:03.247Z","dateModified":"2026-10-10T04:21:16.767Z","description":"OpenAI's Agents API now supports computer use, letting developers spin up a browser and an agent together through a single API call instead of wiring up...","isAccessibleForFree":true,"articleSection":"Collections","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Collections","logo":"https://media.daily.dev/image/upload/s--fk_6ycEi--/f_auto,q_auto/v1780996001/logos/collections?_a=BAMAMiWQ0","url":"https://daily.dev/sources/collections"},"commentCount":1,"discussionUrl":"https://daily.dev/posts/openai-s-agents-api-adds-computer-use-with-a-hosted-browser-96gc6dvzo","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":1},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":1}],"keywords":"ai-agents,openai","timeRequired":"PT2M"}
{"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Collections","item":"https://daily.dev/sources/collections"},{"@type":"ListItem","position":3,"name":"OpenAI's Agents API adds computer use with a hosted browser"}]}
{"@context":"https://schema.org","@type":"WebPage","@id":"https://daily.dev/posts/openai-s-agents-api-adds-computer-use-with-a-hosted-browser-96gc6dvzo","comment":[{"@type":"Comment","text":"oh that’s good","datePublished":"2026-10-07T09:34:22.774Z","url":"https://daily.dev/posts/96Gc6DVzO#c-JTWnQ229y","author":{"@type":"Person","name":"Yosinhon Botirov","url":"https://daily.dev/yosinhonbotirov","image":"https://lh3.googleusercontent.com/a/ACg8ocK38H7UMLhiwxSI39FsR7uS4ghJ42mye4sl7HO_BeZjaYmWfw=s96-c"}}]}
{"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/openai-s-agents-api-adds-computer-use-with-a-hosted-browser-96gc6dvzo#faq","mainEntity":[{"@type":"Question","name":"What does the computer use feature in OpenAI's Agents API actually do?","acceptedAnswer":{"@type":"Answer","text":"It lets developers spin up a browser and an agent together through a single API call, rather than manually wiring up browser automation for an agent separately. The capability to click through a browser isn't new for agents generally, but bundling the browser and agent into one hosted call changes the setup work required. OpenAI has not published detailed documentation on how the hosted browser behaves in practice. Developers tracking agent tooling changes like this follow practical breakdowns on daily.dev."}}]}
```

