Audience: practitioners · Evidence class: mixed
Last reviewed: April 2026.
Computer-use and browser agents operate on GUI surfaces (DOM, pixels, or accessibility trees) rather than pure APIs. Evidence tags follow the Benchmark and Evidence Policy. Entries that are marketing-only or have no active development were removed in this phase.
| Agent | Description | Evidence |
|---|---|---|
| OpenAI Operator | ChatGPT autonomous web agent; human checkpoints; built on the CUA (Computer-Using Agent) model. | [official] |
| Claude Computer Use | Anthropic desktop/browser control via screenshots and tool loop. | [official] |
| Claude for Chrome | Anthropic browsing agent running inside Chrome. | [official] |
| Google Project Mariner | Gemini browser agent with multi-task execution in the user’s browser context. | [official] |
| ChatGPT Atlas | OpenAI’s AI-native browser with Agent Mode. | [official] |
| Dia Browser | AI-native browser from The Browser Company (acquired by Atlassian). | [official] |
| Tool | Description | Evidence |
|---|---|---|
| Browser Use | OSS browser agent library with DOM + vision hybrid; widely embedded in other agent stacks. | [official] |
| Skyvern | Vision-driven browser automation using multimodal LLMs for navigation without coded selectors. | [official] |
| UI-TARS | ByteDance open-source native GUI agent model + desktop app for end-to-end computer use. | [official] · [benchmark] paper |
| Agent S2 (Simular) | OSS compositional GUI automation framework with generalist + specialist models. | [official] · [benchmark] |
| Browserbase | Cloud browser infrastructure for agents; headless Chrome at scale with session persistence. | [official] |
| Amazon Nova Act | AWS browser automation research preview aimed at enterprise reliability. | [official] |
| Playwright MCP | Official MCP server wrapping Playwright for agent-driven browser automation. | [official] |
| Benchmark | Description | Evidence |
|---|---|---|
| OSWorld | Real computer environments benchmark for multimodal agents on open-ended tasks. | [official] · [benchmark] |
| WebArena / VisualWebArena | Reproducible web agent benchmark on real-website snapshots. | [official] · [benchmark] |
| WindowsAgentArena | Benchmark for Windows desktop agents across real applications. | [official] · [benchmark] |