Comparison

v3 vs v4: Choosing a Computer Use Model on the API

Sarah Chen||5 min
+K

You want an API that drives a real desktop, not just a thin wrapper around existing APIs. Coasty Computer Use API gives you vision-based agents that see the screen and act like a human. The API offers two model flavors via the cua_version parameter. v3 is a step-by-step driver suitable for interactive tasks. v4 is autonomous with a built-in pass/fail verifier. This post shows how they differ, when to pick each, and how to call POST /v1/runs with the right cua_version.

How it works

Task runs on the server drive an agent to completion. You POST to /v1/runs with a machine_id, the task text, and cua_version set to either "v3" or "v4". The API provisions a cloud VM, runs a vision model, streams events via Server-Sent Events, and reports final status. v4 adds a verifier that can mark a run as passed or failed without human intervention. Both are billed $0.05 per agent step.

bash
curl -X POST https://coasty.ai/v1/runs \
  -H "X-API-Key: $COASTY_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "machine_id": "vm-12345",
    "task": "Open Chrome, navigate to https://coasty.ai/docs, and print the title",
    "cua_version": "v4",
    "max_steps": 100,
    "deadline_seconds": 3600,
    "on_awaiting_human": "pause"
  }'

v3: Guided, step-by-step automation

  • cua_version: "v3"
  • Agent emits actions and waits for status updates
  • Each agent step costs $0.05
  • Best for tasks that need human-like interaction but can be paused
  • You can resume a run via POST /v1/runs/{id}/resume

Set cua_version: "v3" for guided, interactive automation.

v4: Autonomous runs with verification

  • cua_version: "v4"
  • Agent runs autonomously and verifies pass/fail internally
  • You still get the same per-step billing at $0.05 per agent step
  • Ideal for unattended tasks where you want a clear pass/fail signal
  • Use the same /v1/runs endpoint, just swap the version string

Where this beats brittle automation

Traditional automation relies on brittle selectors, hardcoded coordinates, and fragile API wrappers. It breaks when UI changes or when an app does not expose an API. A computer use agent sees the screen, interprets context, and performs clicks, keystrokes, and typing. It works across browsers, terminals, and desktop apps. v4 adds verification so you know if the job succeeded without inspecting log files or screenshots.

Pick v3 when you want to guide the agent step by step. Pick v4 when you want autonomous verification. Both run on the same /v1/runs endpoint and cost $0.05 per agent step. Start building robust computer use agents at https://coasty.ai/developers.

Want to see this in action?

View Case Studies
Try Coasty Free