Choosing between CUA v3 and v4 determines how you run agents on the Coasty Computer Use API. v3 is a simple, stateless model perfect for controlled task steps. v4 adds an autonomous mode with a built-in pass/fail verifier and a higher max_steps budget. Both are billed $0.05 per agent step, and both use the same POST /v1/runs endpoint. Pick v3 for predictable, step-by-step control. Pick v4 when you want the server to drive the agent to completion with automatic verification.
How it works
You submit a task run with a cua_version field. v3 requires you to stream events and stop the run when done. v4 uses the same endpoint but enables the verifier and lets the server decide when to succeed or fail. The response includes the run status and optional events. You can stream events with GET /v1/runs/{id}/events. States include queued, running, awaiting_human, succeeded, failed, cancelled, and timed_out. v4 can automatically succeed or fail based on the verifier.
# Example: run a CUA v4 task with a pass/fail verifier
export COASTY_API_KEY="your_key_here"
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": "Navigate to https://example.com and click the first link",
"cua_version": "v4",
"max_steps": 50,
"deadline_seconds": 300,
"on_awaiting_human": "pause",
"system_prompt": "You are a helpful assistant that navigates web pages and locates links."
}' | jq .Key differences between v3 and v4
- Both models are billed $0.05 per agent step on POST /v1/runs.
- v3 is stateless, controlled, and best for step-by-step workflows.
- v4 is autonomous with a built-in pass/fail verifier.
- v4 can automatically succeed or fail based on the verifier output.
- v3 requires you to poll /v1/runs/{id} and stop the run when done.
- v4 lets the server decide success or failure and can resume after human approval.
- Both use the same machine_id and task fields.
- Both support max_steps and deadline_seconds to bound execution.
Use cua_version: "v3" for controlled steps and cua_version: "v4" for autonomous execution with a verifier.
Where this beats brittle automation
Traditional automation relies on brittle selectors like XPath or CSS identifiers that break when UI changes. The computer use API lets the agent see the screen, understand context, and act like a human. v3 and v4 models navigate real desktops, browsers, and terminals, not just API calls. This means you can handle dynamic layouts, unexpected elements, and multi-step workflows with a single task definition. You can also use /v1/ground to map screenshots to coordinates for precise clicks, and /v1/parse to convert PyAutoGUI code into structured actions.
Start with v3 for predictable, step-by-step control and move to v4 when you need autonomous execution with verification. Build robust, human-like automation with the computer use API and get your key at https://coasty.ai/developers.
Want to see this in action?
View Case Studies