Anthropic has been quick to reiterate when it comes to adding features to its apps, both on mobile and desktop. All platforms offer most features, although a few are reserved for the desktop application.
Perhaps the main feature is excellent support for MCP (Model Context Protocol). The protocol was designed by Anthropic, but has been widely adopted. That said, there are still many user-facing AI services that don’t do it supports MCP, and that makes a big difference in usability. Native MCP connectors connect Claude directly to Slack, Google Drive, Microsoft 365, Notion, Jira, and more, meaning it can extract context from a company’s actual systems rather than relying on what you paste into a chat window. You can also add custom connectors for any services that support MCP cloud. MCP isn’t just useful for businesses or professionals: connectors for things like Expedia and Gmail mean you can let Claude manage more of your personal life, if you want.
Claude is also great on large pop-ups. Paid plans support popups of up to 1 million tokens, meaning you can insert an entire codebase, an entire application architecture, or a stack of long PDFs, and Claude will reason through the whole thing in a single pass. The ~400,000 token pop-up offered by many other services is, to be honest, more than enough, but if you’re running large projects the extra headroom could definitely come in handy.
This ability fits directly into the coding workflow, which is one of Claude’s strongest areas. Although Claude Code has long been integrated into other IDEs like VS Code, I have largely used it directly in the Claude application. I even built my new website, BandicootLab.com, with it.
Coworking is another feature that sets Claude apart from a standard chatbot, described by Anthropic as bringing the technology behind Code to productivity workflows. In Cowork mode, Claude runs semi-autonomous agentic tasks, including on files on your computer and on any connectors you have configured. It can also work with a Chrome browser extension to handle web browsing tasks. It’s not perfect, and giving an AI this much access to your machine takes some getting used to. But it works, and it works well enough to change how you use the tool.
Then there are artifacts, which allow Claude to natively host interactive apps, dashboards, and standalone calculators linked to your own information. Describe a tool you want, and Claude builds and hosts it — no deployment, no configuration.
The shortcomings are just as notable as the strengths, especially if you’re a user used to having access to things like image generation tools. Claude has no native image generation. For this, ChatGPT is simply the best all-in-one toolkit. You can add image generation with a connector, but you’ll probably have to pay for it. Additionally, it’s worth noting that Claude still has visual tools, like the ability to create SVGs, graphics, and designs with the Claude Design beta.
Voice support is also somewhat limited compared to some competitors. Voice commands work well, but you can’t use your MCP connectors by voice, and it was much more likely to bring up false information, apparently because it prioritizes speaking speed too much. I don’t mind waiting a few seconds for reliable information: if I use voice, it’s because my hands are busy or I’m driving, not because I want to chat.
As of this writing, Claude’s models include Haiku 4.5, Sonnet 5, Opus 4.8, and the recently reinstated Fable 5, in order of model strength. High-end models are only available on paid Claude plans. These models have always been ranked at the top of AI benchmarks. For most uses, the flagship model is Opus 4.8 (Fable 5 will only be available via API). I found it very reliable, but to be honest, all mainstream AI services these days have equally reliable models (Gemini 3.1 Pro and GPT-5.5 are also excellent).
