v3 vs v4: Choosing a Computer Use Model on the API
You need an agent that can see a screen and click, type, and scroll like a human. The Coasty Computer Use API lets you pick a CUA version. v3 is stateless and works well for interactive tasks. v4 adds autonomy and a built-in pass/fail verifier so the server drives the agent to completion. Both charge the same $0.05 per agent step. The right choice depends on how much control you want and whether you need verification.
What is CuaVersion v3
CuaVersion v3 is the default for POST /v1/runs. It is stateless and flexible. The client sends a task and a cua_version field set to "v3". The server returns actions for each step. You capture the screen, call predict, get actions, and repeat until the status is "done". This works well for tasks where you want to loop capture, predict, act yourself.
What is CuaVersion v4
CuaVersion v4 is an autonomous mode with a pass/fail verifier. You set cua_version to "v4" in POST /v1/runs. The server runs the entire trajectory to completion. It returns a final status of "succeeded" or "failed". This removes the need for you to manage the capture-predict-act loop. The server handles vision, reasoning, and verification in one flow.
Choosing between v3 and v4
- ●Use v3 for interactive tasks where you want full control. Example: a custom UI wrapper or a multi-step script you manage.
- ●Use v4 for end-to-end automation with automatic verification. Example: a task that must succeed or fail on its own.
- ●Both versions charge $0.05 per agent step. The difference is control and verification, not price.
- ●v4 supports max_steps, deadline_seconds, and on_awaiting_human options. v3 does not have these constraints built into the run.
- ●v4 is ideal when you want to offload the capture-predict-act loop and focus on task design.
Pick v3 for full control and v4 for autonomous verification, both at $0.05 per step.
Where this beats brittle automation
Traditional automation relies on selectors, XPath, or API bindings. If a UI changes, your scripts break. The Coasty Computer Use API uses vision. The agent sees the screen like a human and acts on what it perceives. This makes your agents more resilient to layout changes and hidden elements. v4 goes further by verifying success automatically, so you do not need custom checks for every task.
#!/bin/sh
# Request a v4 autonomous task run
# Reads COASTY_API_KEY from environment
COASTY_API_KEY=$(cat "$HOME/.coasty_key")
RESULT=$(curl -s -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://example.com, click the first link, and close Chrome",
"cua_version": "v4",
"max_steps": 50,
"deadline_seconds": 300,
"on_awaiting_human": "pause"
} ')
echo $RESULT
# Sample output
# {
# "id": "run-abc123",
# "status": "running",
# "cua_version": "v4"
# }Next steps
Start with a v4 run to see how autonomous verification works. Then switch to v3 for more granular control. Build agents that see, reason, and verify end-to-end. Get your API key at https://coasty.ai/developers and start using the Computer Use API today.