v3 vs v4: Choosing a Computer Use Model on the API
Most computer use APIs give you raw actions. You must build the loop, wait for UI updates, and handle failures yourself. The Coasty computer use API lets you choose how much autonomy you want. The cua_version field tells the server which model to use. v3 is a fully controllable step-by-step model. v4 is an autonomous model that uses a pass/fail verifier to complete tasks without you stepping in. This post shows how each version works, what it costs, and when to use them.
How it works
You start a task run by POSTing to /v1/runs. The request includes machine_id, task, and cua_version. cua_version is a string. v3 is the default. Use "v3" for step-by-step control. Use "v4" for an autonomous agent with a built-in verifier. The server runs the model on a remote desktop, browser, or terminal, and streams events back over Server-Sent Events. You can also use the stateful API with /v1/sessions and /v1/sessions/{id}/predict for persistent trajectory memory.
curl -X POST https://coasty.ai/v1/runs \
-H "X-API-Key: $COASTY_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"machine_id": "vm-123",
"task": "Open Chrome and go to coasty.ai",
"cua_version": "v4",
"instructions": "Do not close the browser",
"max_steps": 50,
"deadline_seconds": 120,
"on_awaiting_human": "cancel"
}'cua_version v3 details
- ●v3 is the default value for cua_version.
- ●The server sends actions to your application via Server-Sent Events.
- ●You must wait for the next screenshot, predict the next action, and call /v1/runs/{id}/cancel or /v1/runs/{id}/resume to control the run.
- ●You can also use the stateful API with /v1/sessions/{id}/predict for persistent trajectory memory.
- ●Billed $0.05 per agent step for task runs.
cua_version v4 details
- ●Set cua_version to "v4" for an autonomous agent.
- ●v4 uses a pass/fail verifier to decide if a task is complete.
- ●The server handles the loop and streams events until the verifier reports success, failure, or a timeout.
- ●You can still cancel or resume a task run.
- ●Billed $0.05 per agent step for task runs.
Pick v3 when you need tight control over every action. Pick v4 when you want the agent to finish tasks autonomously with a built-in verifier.
Where this beats brittle automation
Traditional automation relies on brittle selectors, hardcoded X paths, and fragile APIs. UI changes break everything. The computer use model on the Coasty API sees the screen like a human does. It reads instructions, interprets layouts, and clicks or types based on what it sees. v3 lets you validate every step. v4 lets the model reason to completion. This means fewer flaky tests and more reliable workflows for browsers, desktop apps, and terminals.
Use cua_version "v3" for step-by-step control and debugging. Use cua_version "v4" for autonomous task execution with a pass/fail verifier. Build robust browser, desktop, and terminal automation with the Coasty computer use API. Get your API key at https://coasty.ai/developers.