Claude AI Review 2026: Is Claude Pro Still Worth $20 a Month?

I have used Claude as my main writing and coding tool for over a year, and I pay for it with my own money. This review has been live since June 2026. I have now rewritten it twice: once because two of the criticisms it made of Claude stopped being true, and again today because Anthropic renamed a plan, halved what the top plan promises, and quietly took a model away from the free tier.

That is an uncomfortable way to open a review, so let me be specific about it rather than quietly editing and hoping nobody noticed.

What this review got wrong

The earlier version told you Claude could not run code, and sent you to ChatGPT if you wanted a tool that writes code, runs it, and fixes its own errors. That was wrong. Claude has executed code inside claude.ai since the analysis tool shipped on October 24, 2024, and Anthropic noted on November 5, 2025 that it is being replaced by broader code execution that also generates downloadable spreadsheets, CSVs, and visualisations.

It also told you voice mode was basic, closer to dictation than conversation, and that ChatGPT's was in a different league. Also wrong now. Voice mode is in beta on every plan including Free, across mobile, desktop, and web. There is a hands free mode where Claude listens continuously and replies at natural pauses, the models you get in voice are the same ones your plan gives you in text, and Claude can use connected tools like Gmail and Google Calendar mid conversation. On Free you can connect one of them; paid plans connect more.

I am flagging both because almost every Claude review still repeats them. If you have been avoiding Claude for either reason, the reason has expired.

This round I am correcting myself on four more things, all of them mine rather than Anthropic's: I said the free tier included the Opus family, Claude Code and Cowork, and it includes none of the three. I said Max's second weekly cap applied to Sonnet, and it applies to Fable. I called Enterprise custom priced when it now has a published price. And I described a contradiction in Anthropic's own documentation that has since been fixed, which I have rewritten rather than deleted, because a review that only ever reports other people's corrections is not being straight with you.

The third criticism I made was that Claude cannot generate images. I have softened that one, because I went looking for a page where Anthropic states it and could not find one either way. More on it below.

Claude is not one app any more

This is the biggest change since I first wrote this review, and it is the thing most comparisons miss because they open claude.ai, see a chat box, and describe a chat box.

Anthropic now sells a lot of products under one subscription. There is the chat app. There is Claude Code, which works in your terminal. There is Cowork, an agent that does delegated work across your files and accounts. There is Claude Design, which arrived on April 17, 2026 and builds prototypes, slides, and one pagers. Since I last updated this review the list has grown again: Slides, Docs, Science, Dashboards and Motion in beta, Claude in Chrome, and Claude Security on Enterprise. I have not tested all of them and I am not going to pretend otherwise.

Your subscription is a budget you spend across all of it. That matters for the buying decision, because "is $20 worth it for a chatbot" is the wrong question now. The question is whether you will use more than one of these.

The models you get, and the Fable catch

This is where I had it wrong, so start here. Free gives you Haiku and Sonnet. It does not give you Opus. Anthropic's pricing page marks Opus as a plain "No" on the free column, and I wrote the opposite for two months.

Paid plans from Pro upwards give you Haiku, Sonnet and Opus. Between those plans what changes is how much you can use them, not which ones appear in the picker.

Fable is the exception, and it is the one piece of Anthropic's pricing I would call genuinely confusing. Fable is not available on Free. On Pro it exists, but it does not come out of your plan allowance: it runs on pay as you go usage credits, so using it costs money on top of your $20. On Max it is included, and you can spend up to 50% of your weekly limit on it before you need credits or have to switch models.

So Pro does not really get you Anthropic's top model. It gets you the option to buy it separately. I would rather that were said plainly on the pricing page than sit in a support article.

One honest note on version numbers. The pricing page now names Fable 5.1 directly, at $10 per million input tokens and $50 per million output, but it still lists the Haiku, Sonnet and Opus families without versions. I am not going to assert a precise version for what you will see in your model picker today. Open the picker and look. For how the families compare against OpenAI's current lineup, I went through that separately in GPT-5.6 vs Claude 5.

What the free tier covers

The free tier is better than its reputation, though not as good as I told you in August. No credit card. What Anthropic lists on it: chat on web, desktop and mobile, web search, file creation, running code, memory across conversations, connecting your apps and tools, Artifacts, and Claude Design, Slides and Docs. Voice mode is included too, in beta, on every plan.

What you do not get, and this is the part I got wrong: no Opus, no Claude Code, and no Cowork. Anthropic lists Claude Code as a Pro benefit. You also do not get Fable, the higher usage allowance, or the priority queue when Anthropic is busy, and you are held to a five hour session limit that resets on a rolling basis.

So the free tier is a real trial of the chat app. It is not a trial of the terminal agent or the delegated work agent, which are the two things that make the subscription interesting.

For anyone curious about Claude rather than committed to it, this is a real trial rather than a demo. You will know within a week whether the writing quality is worth paying for.

Pro at $20, and the annual price nobody mentions

Pro is $20 a month, or $17 a month if you pay annually, which Anthropic bills as $200 upfront. That annual option is easy to miss and saves $40 a year.

What changes at Pro is headroom and access. The ceiling that interrupts free users moves far enough out that a normal working day does not hit it. Pro is also where Claude Code starts, along with Projects, handing off and scheduling tasks, Claude Science, Claude Dashboards in beta, Claude in Chrome and Microsoft 365, and the priority queue when traffic spikes. Enterprise search is not on Pro; that one starts on Team, and I had it in the wrong column before.

Anthropic also doubled Claude Code's five hour rate limits on May 6, 2026 for Pro, Max, Team, and seat based Enterprise, and removed the restriction that used to throttle Claude Code during peak hours on Pro and Max. If you tried Claude Code on Pro before then and found it ran out too fast, that specific complaint has been addressed.

The reason those limits could rise is worth a sentence, because it is unusually concrete for this industry. Anthropic signed an agreement with SpaceX for the entire compute capacity of the Colossus 1 data centre, which it described as over 300 megawatts and more than 220,000 NVIDIA GPUs coming online within a month, and said Pro and Max subscribers were the intended beneficiaries.

Max and Ultra, and the rename you probably missed

If you read this review before November, the plan names in it are gone. There is no Max 5x and no Max 20x. There is Max at $100 a month, and above it Ultra at $200. Anthropic's help centre puts it plainly: the Max 20x plan is now called Ultra, and existing Max 20x subscribers moved across automatically with no price change.

The rename came with a change to the headline number, which is the part worth noticing. Max 20x was sold as twenty times Pro's usage. Ultra is documented as ten times Pro's usage per week. Same price, same subscribers, half the stated multiple. I cannot tell you whether the real allowance moved or whether the old number was never accurate, because Anthropic has not published the arithmetic behind either figure.

Max is the clearer of the two: five times Pro's usage, every five hours and every week, and Fable included up to half your weekly limit instead of billed as credits. If you use Fable regularly that alone can justify the gap between Pro plus credits and Max.

Ultra has a stranger shape. Its selling point is that there is no five hour limit at all, so a long session never stops. Except that this depends on your account being verified, Anthropic verifies accounts itself using what it calls factors like your account and payment history, and it says explicitly that verifying your phone number at checkout does not count. Until that happens, a temporary five hour limit applies, and the way you find out which side of it you are on is to open Settings and look for a session bar. A $200 plan whose main benefit arrives on a timetable you cannot see is not a good look.

Both plans now bundle API credits for building on Anthropic's platform: $100 a month on Max, $200 on Ultra. If you were already paying for both a subscription and API access, that is a real saving and it is new since August.

How the limits actually work

This is the part I find hardest to explain to people, and I think that is Anthropic's fault rather than theirs.

Every plan runs on a rolling five hour session window, except verified Ultra accounts, which have none. Paid plans add a weekly cap on top. Max and Ultra each carry two separate weekly caps, one across all models and one that applies to Fable alone. I previously wrote that the second cap was for Sonnet. It is for Fable, and that is my error rather than a change at Anthropic's end. When you run out, you can buy usage credits or a limit reset instead of waiting.

I have described daily limits in the past and that was never quite right. The five hour window is the one that bites during a working day: a heavy morning can slow your afternoon even though you are nowhere near a weekly cap.

The credits system is fair enough in principle, and I would rather have the option than be locked out. But between session windows, weekly caps, a separate Fable cap, Fable billed differently on every plan, account verification deciding whether Ultra's main promise applies to you, bundled API credits, and top up bundles, working out what you will actually spend in a month takes real effort. That is my main criticism of Claude in 2026, and it is a pricing criticism rather than a product one.

Artifacts and Projects still carry the daily work

Artifacts is the feature I would miss most. When Claude produces code, a web page, a React component, a table, or an SVG, it opens in a live panel next to the conversation and runs. Edit the code, watch the preview update. Every artifact gets a URL you can send to someone without a Claude account, which is genuinely useful for showing a client a prototype.

Projects solves the other everyday annoyance, which is re explaining your context every time you start a conversation. A project is a persistent workspace holding files Claude always has access to, custom instructions for that work specifically, and the history of every conversation inside it.

The difference from ChatGPT's memory is scope. ChatGPT keeps one picture of you and applies it everywhere. Projects keeps a separate knowledge base per piece of work, so my writing setup and my code setup do not contaminate each other. If you juggle several clients, this is the feature that changes your week.

Cowork is the interesting one

Cowork is the piece I am least settled about, which is why I want to describe it carefully rather than sell it.

The idea is delegation rather than conversation. You hand Claude a task and it works across your files, calendar, email, messaging apps, the web, and connected tools until it finishes, escalating to you when it reaches a judgement only you can make. Anthropic's example is scheduling Monday client prep for 6am, so Claude reads the email threads, transcripts, and recent news, builds the briefing document, and leaves the follow up email drafted but unsent.

Two details worth knowing. Scheduled tasks run with no device online, because the sessions are hosted on Anthropic's servers rather than your laptop. And capability differs by platform: on desktop Claude can use your local files and browser, while web and mobile have narrower access.

Anthropic brought it to web and mobile on July 7, 2026, starting with Max. I have used it for scheduled research summaries and it does the boring parts well. I would not yet hand it anything where a wrong call is expensive, and the fact that it asks before those calls is the design decision I most approve of.

Writing quality

This is still why I pay.

For long pieces, Claude produces prose that needs less repair than anything else I have tried. Fewer hollow transitions, less padding, more willingness to state something plainly instead of hedging it into mush.

The specific thing it does better than its competitors is hold a complicated brief. Tell it the voice, the audience, the terminology, the phrases to avoid, and the length, and it will still be obeying all of those in paragraph forty. Most tools start shedding constraints as the text gets longer. For anything with a real brief attached, that is the difference between editing and rewriting. I described the workflow I actually use in how to write blog posts faster with AI.

I want to be fair to the competition. For short work, emails, captions, product descriptions, the gap is small enough that I would not switch tools over it. It shows up at length. If you want the direct comparison, Claude vs ChatGPT goes through it properly.

Coding

Claude follows a specification more closely than most competitors. Write a thorough spec and you are less likely to get something that looks correct but misread requirement four. It still makes mistakes, as they all do, but the miss rate on carefully described tasks is lower.

Code execution changes the loop meaningfully now that it exists in the chat. You can have Claude write something, run it, see the actual error, and fix it, without leaving the conversation. That was the gap I used to point at, and it has closed.

For real project work I use Claude Code in the terminal rather than the browser, because it reads and edits files in place and runs tests. Pro covers it within limits. I wrote up the workflow in how to use Claude AI for coding.

If you would rather the assistant lived in your editor than your terminal, that is a separate subscription with its own metering. Cursor vs GitHub Copilot compares what those two actually cost once agent usage is counted.

Where Claude still falls short

No raster image generation that I can find. In a year of daily use Claude has never produced a photograph, a logo or an illustration for me, and the feature does not appear anywhere in Anthropic's plan comparison. I went looking for a page that states the limitation outright and there is not one, so treat this as what I have observed rather than a documented restriction. What Claude does do is vector and code based: SVG, charts, diagrams, interactive artifacts, and through Claude Design, prototypes and slides that export to PPTX, PDF, HTML, and Canva. There is also a workaround worth knowing, which is that connecting an image generation MCP server lets you prompt for images inside a normal chat. Still, if visual assets are core to your work, budget for a separate tool. I keep a list of free AI image generators for exactly this.

Pricing is hard to predict. Covered above, and it is my strongest complaint.

Hallucination has not gone away. Claude is better than most at saying it is unsure, and it will tell you to verify a figure. It also still states wrong things confidently. Anything that matters, check yourself. That applies to this review too: I have listed sources so you can.

The third party ecosystem is smaller. Claude has Projects, connectors, MCP, and a plugin marketplace inside Cowork, but there is nothing the size of ChatGPT's store. If you depend on community built tools, that is a real gap.

Anthropic's documentation moves under you. The context window was my example here: in August the pricing page said 200k for Pro while the help centre described up to 1M on newer models, and I used the gap as evidence of carelessness. That particular gap has closed. The pricing page now says up to 1M for every plan, varying by model, which is both consistent and vague enough that you still cannot plan around it. The broader complaint stands, and this review is the evidence: in ten weeks a plan was renamed, a headline multiple halved, and a free tier quietly lost a model family, none of it announced anywhere I could find.

Who should buy what

Free is enough if you use AI a few times a week, or if you want to find out whether the writing quality lives up to the reputation before paying.

Pro at $20 is right for most working professionals. Writers, developers, anyone who works through long documents daily. Pay annually at $17 if you already know you will keep it.

Max at $100 makes sense in two situations: you regularly exhaust Pro, or you want Fable included rather than buying credits on top. Work out your credit spend on Pro first, because that comparison decides it. The $100 of monthly API credits counts too, if you build anything.

Ultra at $200 is for people running Claude as the centre of their work rather than a tool beside it. Check Settings and confirm your account is verified before you count on the no session limit promise, because that is the whole reason to pay the extra $100.

Stay with ChatGPT instead if you need native image generation, or if the third party ecosystem is what you are buying.

Pricing

Prices and plan contents verified against Anthropic's pricing page and help centre on November 3, 2026. Ultra did not exist under that name when I last checked on August 5, and Enterprise had no published price. This is a product that reprices and renames, so check before you buy.

My verdict

Claude is still the best writing tool I have used, and it is now a better coding tool than it was when I first reviewed it, because the one thing I criticised has been fixed.

What has changed is what you are buying. A year ago Claude was a very good chat app. Now it is a chat app, a terminal agent, a delegated work agent, a design tool and about six betas sharing one usage budget. If you will use two of those, $20 is straightforwardly good value. If you only want the chat box, the free tier is a fair trial of it, as long as you know going in that Opus, Claude Code and Cowork are not part of the deal.

What has not improved is Anthropic's willingness to tell you when the terms change. Writing this update meant reading four help centre articles and a pricing table to discover a renamed plan and a halved usage promise. Keep the receipt for whatever you sign up to.

Rating: 9 out of 10.

The missing point is not what it used to be. Code execution arrived and voice mode became genuinely good, so those complaints are gone. What remains is no native raster image generation, and a pricing structure I cannot predict a month ahead without doing arithmetic across session windows, weekly caps, model specific caps, and credit top ups. The product earned back the point I docked it. The pricing page lost it again.

Try Claude for free. No credit card required.

Verified on August 5, 2026 against Anthropic's pricing page, the Fable plan support article, the usage and length limits article, the analysis tool announcement, the voice mode help page, the Claude Design launch post, and the Cowork web and mobile announcement. Anthropic ships often, so treat the dated claims above as dated rather than permanent.