← Models

Computer Use Preview

OpenAI · Computer Use

OpenAI's Responses API computer-use model for supervised browser and interface-control workflows.

Part of GPT family · Other versions: GPT-5.6 Luna , GPT-5.6 Terra , GPT-5.6 Sol , GPT-5.5 Pro , GPT-5.5 , GPT-Rosalind , GPT-5.4 mini , GPT-5.4 nano , GPT-5.4 Pro , GPT-5.4 , GPT-5.3 , GPT-5.3-Codex , GPT-5.2-Codex , GPT-5.2 Pro , GPT-5.2 , GPT-5-Codex , GPT-5 mini , GPT-5 nano , GPT-5 , GPT-4.1 , GPT-4o mini Transcribe , GPT-4o mini TTS , GPT-4o Transcribe , GPT-4o mini , GPT-4o
Type
multimodal
Context
8K tokens
Max Output
1K tokens
Status
preview
Input
$3/1M tok
Output
$12/1M tok
API Access
Yes
License
proprietary
computer-use browser-control agentic multimodal responses-api preview
Released March 2025 · Updated June 8, 2026

Overview

Freshness note: Model capabilities, limits, and pricing can change quickly. This profile is a point-in-time snapshot last verified on June 8, 2026.

Computer Use Preview is OpenAI’s model-level route for supervised computer-control workflows through the Responses API. It matters because browser and interface control are not just ordinary text generation: the model observes a screen state, reasons about a next action, and emits bounded computer-use actions that an application can execute under a tool policy.

Capabilities

This model class is aimed at workflows where an agent must inspect a page, reason about interface state, and take bounded actions such as clicking, filling fields, navigating, or collecting evidence. That makes it relevant for legacy tools, internal dashboards, QA flows, and operational web workflows where APIs are incomplete or unavailable.

Technical Details

OpenAI’s current model page lists Computer Use Preview with an 8,192-token context window and 1,024 max output tokens. It is available through the Responses API, not Chat Completions or Batch, and is intended to be used with the computer-use tool guide rather than as a general multimodal assistant model.

The model can inspect screenshots or browser states and emit actions, but the runtime design is the real safety boundary. Human approval, allowed-domain controls, action logging, and high-risk-action blocks matter more than the model name.

Pricing & Access

OpenAI’s current model page lists Computer Use Preview pricing at:

  • Input: $3.00 per 1M tokens
  • Output: $12.00 per 1M tokens

Availability is through OpenAI’s Responses API. Teams should assume preview behavior changes and stricter evaluation needs than on stable text-only production models.

Best Use Cases

Use Computer Use Preview for supervised browser automation, internal operations tooling, form-based workflows, and evaluation environments where a reviewer can approve actions before systems of record are changed. It is not a good fit for unsupervised high-risk automation.

Comparisons

  • GPT-5.5 (OpenAI): Stronger general reasoning route; Computer Use Preview is the specialized UI-action model.
  • Perplexity Computer (Perplexity): Packaged browser-agent product for end users, while Computer Use Preview is a model-layer building block.
  • Gemini computer-use surfaces (Google): Comparable browser-control lane, with provider choice usually driven by runtime, policy, and ecosystem fit.