Skip to content
Gallery
Autonomy Product Benchmarking
Models / Products in Scope
Framework Design
Feedback
Operator Testing
Anthropic Computer Use
Amazon Nova Act
Share
Explore

icon picker
Models / Products in Scope

Detailed Model Comparison
OpenAI CUA vs Anthropic CU vs Nova Act
Dimension
Amazon Nova Act
OpenAI Operator / CUA
Anthropic Computer Use
Release & status
 Mar 31, 2025
Research‑preview SDK on wait‑list (US‑only) 
 Jan 23, 2025
Research‑preview inside ChatGPT Pro (US) 
 Oct 22, 2024
Public‑beta feature in Claude 3.5 Sonnet API 
User‑facing abstraction
“Selenium you can talk to” — dev composes many small act() calls; no consumer UI
Chat‑style “do this for me” button in ChatGPT; end‑to‑end goals
Pure developer feature; you embed Claude in your own agent loop
Scope of interface
Web browser only (Chromium via Playwright) 
Web browser (headless Linux VM) 
Full desktop: browsers + terminals + file‑system windows 
Autonomy style
Reliability‑first atomic steps you chain together
One high‑level prompt; CUA plans & self‑corrects
Dev supplies the loop; Claude can plan multi‑step but hands control back each iteration
Core loop
Screenshot → Nova model → JSON list of low‑level actions → Playwright executes
Screenshot → CUA → actions (click/type/scroll) → virtual mouse & keyboard
Same screenshot‑action loop, but dev calls the Claude API directly
Developer tooling / reference
nova‑act Python SDK, Playwright escape‑hatch, HTML replay & optional MP4 video logs 
 No public SDK yet; only ChatGPT UI & REST operator endpoint
Dockerized Ubuntu VM + Streamlit demo; JSON inner_thoughts trace 
Introspection / transparency
Full HTML timeline + video; you can inspect each act() result
Shows high‑level plan in chat; no structured trace
JSON inner_thoughts reveals Claude’s step‑by‑step reasoning
Memory / personalization
None; you manage cookies or user‑data‑dir manually
Persistent user instructions & history in ChatGPT
Stateless unless you build a wrapper
Human‑in‑the‑loop
Recommend monitoring; no built‑in HITL UX
Operator asks for confirmation on risky steps (CAPTCHAs, log‑ins) 
Dev decides when to hand control back (e.g., for captchas)
Licensing & availability
Apache‑2.0; experimental; API keys from nova.amazon.com 
Closed‑source; accessible to $200/mo Pro tier
Available through Anthropic API, Bedrock, Vertex; standard Claude terms 
There are no rows in this table
⁠


Want to print your doc?
This is not the way.
Try clicking the ⋯ next to your doc name or using a keyboard shortcut (
CtrlP
) instead.