v3 vs v4: Choosing a Computer Use Model on the API
You want an agent that interacts with your desktop, browser, or terminal exactly like a human. The Coasty Computer Use API (CUA) gives you two main model versions: v3 for precise task-driven steps and v4 for fully autonomous agents with a pass/fail verifier. Choosing the right CUA version changes how you build, test, and pay for your agents. This guide shows the concrete differences, the real request fields, and a working example for each path.
How it works
Both versions start with a POST /v1/runs request. You must provide a machine_id, a task, and a cua_version. The cua_version field selects the model behavior. v3 runs the agent through discrete steps until it reaches the goal. v4 adds a verifier that evaluates the final state against success/failure criteria. The API returns a run_id and streams events via GET /v1/runs/{id}/events. Billing runs at $0.05 per agent step on both models. You can pause or cancel runs using POST /v1/runs/{id}/cancel or POST /v1/runs/{id}/resume.
curl -X POST https://coasty.ai/v1/runs \
-H "Content-Type: application/json" \
-H "X-API-Key: $COASTY_API_KEY" \
-d '{
"machine_id": "vm-12345",
"task": "Open Chrome, navigate to cozy.coasty.ai/docs, and capture the title.",
"cua_version": "v3",
"max_steps": 20,
"on_awaiting_human": "pause"
}'
v3: precise task-driven agent
- ●Set cua_version to "v3" in the run request.
- ●Each agent step costs $0.05.
- ●You control max_steps and deadline_seconds.
- ●on_awaiting_human lets you pause instead of failing.
- ●Events stream step-by-step progress.
Use v3 when you want predictable, stepwise automation with full control.
v4: autonomous agent with pass/fail verifier
- ●Set cua_version to "v4" for autonomous mode.
- ●The v4 model runs a verifier against the final state.
- ●Verifying success or failure is built into the agent loop.
- ●Billed at the same $0.05 per step price.
- ●Great for end-to-end workflows that need automatic validation.
Where this beats brittle automation
Traditional automation relies on brittle selectors, element IDs, or API mocks. If a UI changes, your scripts break. Coasty's computer use model sees the screen, interprets instructions, and clicks, types, and navigates like a human. With v4, the verifier checks the actual result instead of hardcoded assertions. This lets you build agents that adapt to layout changes, popups, and dynamic content without rewriting selectors.
Pick v3 for fine-grained, controllable tasks. Pick v4 for end-to-end automation with built-in validation. Both run on real machines and bill at $0.05 per step. Get your API key at https://coasty.ai/developers and start building agents that see and act like real users.