Building a computer use agent that can reliably drive desktop apps, browsers, and terminals is easier when you pick the right model. Coasty offers two cua_version values: v3 and v4. v3 is a deterministic, step-by-step model that you control until completion. v4 is autonomous with a built-in verifier that can succeed or fail the task on its own. This post compares the two so you can make the right choice for your use case.
How the API works
You start a task run by POSTing to /v1/runs. The request includes a machine_id, a task, and the cua_version you want to use. The body looks like this. You can also add instructions, a system_prompt, max_steps, deadline_seconds, on_awaiting_human, and a webhook_url. The server bills $0.05 per agent step when the run is in the running state. The response gives you an id and an initial status (queued, running, awaiting_human, succeeded, failed, cancelled, timed_out). You can poll GET /v1/runs/{id} or stream events with GET /v1/runs/{id}/events until the status changes to succeeded or failed.
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 cozy.com, and log in",
"cua_version": "v4",
"instructions": "Use the browser to navigate to the login page",
"max_steps": 50,
"deadline_seconds": 300,
"on_awaiting_human": "pause",
"webhook_url": "https://your-domain/webhook"
}'v3: Deterministic, step-by-step control
- cua_version: v3
- You control the agent step by step
- Status stays running until you send cancel or the task completes
- Useful when you need fine-grained validation after each action
- Billed at $0.05 per agent step while running
Use v3 when you want full control over each step and can manually validate the state after each action.
v4: Autonomous with built-in verifier
- cua_version: v4
- The agent drives the task to completion or failure
- Verifier determines success or failure automatically
- Simpler workflow: you only need to check the final status
- Billed at $0.05 per agent step while running
Use v4 when you want an autonomous agent that can complete the task and you only need to see succeeded or failed.
Where computer use beats brittle automation
Traditional automation relies on brittle selectors and API-only tools. When UI changes, selectors break and maintenance costs explode. Computer use agents see the screen and act like a human. They can handle dynamic layouts, new buttons, and context-aware decisions. On the official OSWorld leaderboard, Coasty’s computer use agent achieved 85.6% task success on in-house tests and 82.81% on the public leaderboard. This reliability comes from vision-based control and stateful trajectory memory, not fragile selectors.
Pick v3 for full control and v4 for autonomous completion. Start building with a free key at https://coasty.ai/developers .
Want to see this in action?
View Case Studies