Claude (Anthropic)

What it actually does

Claude is Anthropic's chat assistant, and the thing it is genuinely best at is long text and code. The 200K token context window is the largest among the major assistants, so a whole codebase, a book manuscript or a stack of contracts fits into one conversation without chopping it into pieces first. Artifacts pulls generated code, diagrams and documents out of the chat into a panel you can keep editing. Projects holds a set of files and standing instructions so you are not re-explaining yourself every morning.

The current model is Claude 3.5 Sonnet, with vision for images and screenshots, and a Computer Use beta that drives a desktop. G2 reviewers, 462 of them, put it at 4.6 out of 5 and keep singling out coding quality and long-form reasoning. We agree with that part. What it does not do is generate images, and its plugin ecosystem and integration list are both smaller than ChatGPT's, which matters if you want the assistant wired into the rest of your stack.

What it costs

There is a free tier with a limited number of messages. Pro is $20 a month, Team is $30 per seat per month, and Enterprise is quoted privately.

The sticker price is not the problem. What $20 actually buys is. Usage is metered on a rolling five-hour window plus a weekly quota, enforced server-side, and the only place you can see where you stand is a usage settings page most subscribers never open. Nothing at checkout tells you how many messages or how many hours of coding work a month contains, because Anthropic does not publish that number in any unit a buyer could plan around. If your work is bursty, you will pay $20 and lose Friday.

Where it falls down

Usage limits, and nothing else is close. Paying subscribers report hitting a wall mid-task and say the caps have been tightened repeatedly and quietly. Users describe burning 25% of a weekly allowance in an hour with no warning before the cutoff.

Ars Technica reported in May 2026 that the changes were "a response to a lot of user frustration about a compute crunch, especially in recent weeks," and that "Anthropic has been testing solutions to reduce demand, like enforcing stricter limits during peak hours." Anthropic also tested removing Claude Code from the $20 Pro plan altogether. Pulling a flagship feature out of a tier people have already paid for, even as a test, tells you how the company weighs your subscription against its compute bill.

The pattern subscribers now describe is a loop: cut the limits, wait for the outrage, restore part of them, then launch a more expensive tier. Trustpilot sits at 1.5 out of 5 across roughly 1,963 reviews, largely people who paid for a month, hit the ceiling and concluded the subscription was unusable for sustained work. G2's 4.6 is the same product rated by people who never reach the wall. Both numbers are honest. They just describe different ends of the quota.

In fairness, Anthropic responded to the backlash by doubling some Claude Code limits and removing the peak-hour reductions. That is a real improvement, and it arrived after public anger rather than before it.

Who should buy it

Writers, researchers and legal professionals working with long documents should buy it, because the output quality is the best in the category and document-sized reasoning is where the context window earns its keep. Developers should buy it too, with eyes open about the ceiling.

Do not buy Pro if your plan is eight hours a day of agentic coding. You will hit the weekly quota, you will not have seen it coming, and the fix is a more expensive tier rather than an apology. Check the usage settings page in your first week and decide from that, not from the marketing.

Do not buy it at all if you need image generation or a deep integration catalogue. ChatGPT covers both and Claude does not pretend to. Pay monthly. The limits have changed enough times that an annual commitment is a bet on a policy Anthropic has not held still on.