Choosing the right CUA version changes how your computer use agent behaves and how robust it stays. v3 is a stateless, explicit task runner. v4 adds autonomy and a built-in verifier. Both cost $0.05 per agent step, but they differ in how they handle verification, retries, and human input. This post compares the two so you can pick the right model for your automation scenario.
How it works
Both versions use the task runs API. You send a POST to /v1/runs with a machine_id, task, and cua_version. v3 returns actions and status until the task finishes. v4 returns a pass or fail status and can automatically retry on failure. The server streams events to GET /v1/runs/{id}/events so you can track progress. States include queued, running, awaiting_human, succeeded, failed, cancelled, and timed_out.
curl -s 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 and navigate to https://example.com",
"cua_version": "v4",
"max_steps": 50,
"deadline_seconds": 600,
"on_awaiting_human": "pause"
}'Key differences
- v3: Stateless task runner. Returns actions and status at each step. Good for explicit step-by-step control.
- v4: Autonomous with pass/fail verifier. Returns success or failure after the task completes. Handles retries and human approval automatically.
- Both versions cost $0.05 per agent step.
- v4 requires a verifier to return true/false for pass/fail.
- on_awaiting_human can be pause, fail, or cancel in both versions.
Pick v3 for fine-grained control, v4 for autonomous verification.
Where this beats brittle automation
Traditional automation relies on brittle selectors and static API calls. If a UI changes or an element moves, your script breaks. A computer use model reads the screen, understands instructions, and acts like a human. It adapts to layout changes, modal dialogs, and unexpected UI elements. This makes your automation resilient to product changes and reduces maintenance overhead.
Use v3 when you want to inspect each action or integrate tightly into your workflow. Use v4 when you need an autonomous agent that verifies results and handles retries. Start building with the Coasty Computer Use API at https://coasty.ai/developers and get your API key today.
Want to see this in action?
View Case Studies