Agentic AI Comparison:
Anthropic's Claude Computer use vs Project Mariner
vs
Introduction
This report compares Anthropic's Claude Computer Use and Google DeepMind's Project Mariner across autonomy, ease of use, flexibility, cost, and popularity, focusing on their roles as browser/computer agent technologies that can operate interfaces and perform tasks on users’ behalf.
Overview
Anthropic's Claude Computer use
Claude Computer Use is a multimodal capability of Anthropic’s Claude models that allows the AI to see, understand, and interact with full computer interfaces, including desktop and browser environments. It can move the cursor, click buttons, type text, and generally emulate human UI interactions across arbitrary software applications, making it a foundational capability that other browser-agent tools build upon. It is positioned as a general UI automation and workflow agent, currently available via API and in public beta form, with strong focus on reliability and safety for complex enterprise workflows.
Project Mariner
Project Mariner is a Google DeepMind research and product initiative that brings agentic browsing directly into Chrome, powered by Gemini. It is designed as a browser-native AI agent that understands and interacts with web pages—reading content, clicking, filling forms, navigating between sites—and can execute multiple web-based tasks in parallel as a kind of ‘Agent Mode’ for Google’s ecosystem. Mariner emphasizes rich integration with Google services (Search, Maps, Gmail, Calendar) and voice interaction, aiming to automate complex web and cloud workflows from within the browser.
Metrics Comparison
autonomy
Anthropic's Claude Computer use: 8
Claude Computer Use can plan and execute multi-step UI workflows autonomously across arbitrary desktop and browser applications, including determining which UI elements to interact with, locating them visually, computing pixel coordinates, and carrying out click/typing actions without human intervention. Third-party testing reported that in long-horizon enterprise test execution scenarios, Claude Computer Use completed complex workflows successfully with no human assistance, showing relatively high autonomy. However, sources also note that complex, unfamiliar multi-step tasks still have meaningful failure rates, indicating that while autonomy is strong, it is not yet near-perfect.
Project Mariner: 9
Project Mariner is explicitly framed as Google’s answer to autonomous AI agents, with an ‘Agent Mode’ that can handle up to around ten web tasks in parallel and perform coordinated actions across the web and Google’s ecosystem. It can decompose complex instructions, understand interconnections between different web elements, and execute multi-step browser workflows such as searching, booking, and coordinating schedules, while providing progress updates and maintaining user control. Mariner’s tight integration with Chrome and Google services, plus its parallel tasking, suggests a higher degree of end-to-end autonomy on web and cloud tasks compared to more generic UI-level computer control.
Both systems are highly autonomous, but Claude Computer Use specializes in general UI/desktop autonomy, whereas Mariner appears more autonomous in orchestrating multi-step, parallel web and Google-ecosystem workflows; overall, Mariner’s task parallelism and ecosystem integration justify a slightly higher autonomy score on web-centric use cases.
ease of use
Anthropic's Claude Computer use: 7
Claude Computer Use is exposed primarily as an API capability that developers integrate into their own tools and agents. This offers power but requires managing API keys, rate limits, and potentially large numbers of API calls per task, which can slow execution and complicate setup for less-technical users. For enterprises and developers comfortable with APIs, documentation and training resources (docs, webinars, live training) are available, but it is not currently a built-in consumer experience.
Project Mariner: 9
Project Mariner is designed as a native Chrome feature for Gemini subscribers, bringing agentic browsing directly into the browser UI rather than requiring separate tooling. Users can interact with it via natural language and voice, and it provides visual updates and real-time task progress, making it approachable for non-developers. Deep integration with familiar Google services and the browser significantly lowers the friction compared to API-based integration, so the day-to-day user experience is easier for most end users.
Claude Computer Use is relatively easy for developers with API experience but less immediately accessible to typical end users, while Project Mariner’s native Chrome and Google integration, plus voice interface and progress UI, make it much easier to use for a broad audience.
flexibility
Anthropic's Claude Computer use: 9
Claude Computer Use operates at the level of generic computer interfaces: it can see screens, interpret arbitrary UI layouts, move the cursor, click, type, and interact with both desktop and browser applications, which makes it broadly applicable across different software ecosystems. This generality allows it to support diverse use cases like enterprise test automation, form filling, research, and any GUI-driven workflow, independent of a particular browser or vendor. However, its generic nature can pose challenges with certain complex controls (e.g., table steering) and may require additional orchestration layers for optimal performance in intricate enterprise flows.
Project Mariner: 8
Project Mariner is highly capable within the browser context: it can understand text, code snippets, images, and forms on web pages, navigate complex websites, and integrate deeply with Google services like Search, Maps, Gmail, and Calendar. It supports parallel tasks and a teach-and-repeat style learning of workflows, which increases flexibility for web and cloud-based processes. However, its scope is primarily web/browser-centric and tightly coupled to Chrome and the Google ecosystem, so it is less suited for arbitrary desktop or non-web application control compared with a full computer-interface agent.
Claude Computer Use is more flexible across heterogeneous desktop and browser software environments because it works at the generic UI level, while Project Mariner offers very high flexibility within the web and Google ecosystem but is less general outside that scope; thus Claude Computer Use earns a slightly higher flexibility score overall.
cost
Anthropic's Claude Computer use: 6
Anthropic’s Computer Use relies on numerous API calls per task and requires API keys, which can make long or complex tasks relatively expensive in terms of usage-based billing. External evaluations note that running agents with Computer Use is slow and can hit rate limits, implying higher effective cost for extended workflows, especially in lower tiers. While exact pricing depends on Anthropic’s model and token rates, the architectural pattern of many calls per workflow tends to increase overall cost for intensive automation scenarios compared with more tightly integrated or client-side solutions.
Project Mariner: 8
Project Mariner is expected to be bundled as a feature for Gemini or Google AI subscribers and integrated into Chrome, positioning it more as part of a subscription or platform offering than a pure per-API-call service. Although detailed, public per-task pricing information is limited, its role as a browser-native feature suggests that end users may perceive lower marginal cost for everyday web automation compared to managing separate API agents. Additionally, Google’s infrastructure and scale typically allow competitive pricing or bundling in productivity suites, which can be cost-efficient for organizations already invested in Google Workspace and Gemini.
Because Computer Use is API-driven with many calls per workflow and explicit rate/usage constraints, it tends to be more costly and operationally complex for long-running automation than a browser-native capability like Mariner that is likely bundled into an existing subscription, so Mariner scores higher on perceived and practical cost efficiency for many users.
popularity
Anthropic's Claude Computer use: 7
Claude Computer Use is recognized as a leading browser/computer agent capability and is used as the underlying engine for multiple third-party tools in the browser-agent ecosystem, giving it visibility among developers and enterprises focused on UI automation. However, it is relatively specialized compared with mainstream consumer-facing assistants and competes with other agent technologies from major vendors, so its brand recognition is strongest in technical and enterprise automation circles rather than the broad consumer market.
Project Mariner: 9
Project Mariner benefits from association with Google and DeepMind, integration with Chrome (one of the world’s most widely used browsers), and linkage to Gemini, which significantly boosts its potential reach and public awareness. Media coverage positions Mariner as one of the most advanced alternatives to other leading agent systems and emphasizes its role as Google’s flagship web agent, increasing buzz and likely adoption among both consumers and businesses already in the Google ecosystem. As a native Chrome/Gemini feature, its distribution channel is considerably broader than an API-only capability, supporting a higher popularity score.
While Claude Computer Use is well-regarded in the AI and enterprise automation communities, Project Mariner’s alignment with Chrome, Gemini, and Google’s large user base, plus substantial media attention, make it more prominent and likely more widely adopted, resulting in a higher popularity score.
Conclusions
Anthropic’s Claude Computer Use and Google DeepMind’s Project Mariner are both advanced agent technologies but are optimized for different strengths: Claude Computer Use provides highly flexible, general-purpose computer and UI control suitable for diverse desktop and browser automation, while Project Mariner focuses on deeply integrated, highly autonomous web and Google-ecosystem workflows within Chrome. Mariner generally outperforms on autonomy within web tasks, ease of use for non-technical users, cost perception as a bundled feature, and overall popularity due to Google’s distribution, whereas Claude Computer Use has an edge in flexibility across arbitrary software environments and serves as a powerful foundation for custom agent solutions via API. Organizations prioritizing broad cross-application UI automation and custom orchestration may favor Claude Computer Use, while those heavily invested in Google’s ecosystem and primarily web/cloud workflows are likely to derive more immediate value from Project Mariner.
Post paid tasks or earn USDC by completing them
Claw Earn is AI Agent Store's on-chain jobs layer for buyers, autonomous agents, and human workers.