Claude

What to weigh: whether you need an assistant for long-haul work.

Who it is for. Anyone working with long documents, code and text. Claude makes things up less often when pulling out facts and holds a given tone longer - that is how practitioners explain businesses moving over from ChatGPT (Habr, Mar 25, 2026). Context on paid plans runs up to a million tokens, several thick books in one conversation.

Who else. Anyone who wants the top-ranked model without paying extra. Opus 5.5 came out on Sep 22, 2026 and scored 58 points in the Artificial Analysis index at maximum effort, the highest in the index. It is included in Pro for $20, while Fable 5.1 with its 53 points is available on Pro only for extra credits.

Who it is not for. Those who need images and video: Claude doesn't make them, and here ChatGPT and Gemini are stronger.

Who else it is not for. For people who count messages. The vendor does not publish fixed limits: there is a five-hour window plus a weekly cap, and practitioners cannot find a counter or warning before hitting the limit.

Choose between it and ChatGPT: see section 7.

What vibecoding isStarting from scratch: how to set tasks for agents with no coding experience

Three layers: chat with Anthropic models, the Cowork agent in your folders, and an extension that clicks in the browser.

Chat. Questions, web search, analysis of uploaded files, voice mode, memory across conversations and projects with a shared document base. You pick the model: the free plan has Sonnet 5.5 and Haiku 4.5, and paid plans add Opus 5.5 and Fable 5.1 with a million-token context. A long answer arrives as an artifact, a separate window with a document, table, code or an interactive page.

Cowork. You name the goal and the folder, and the agent reads the files itself, builds a plan and brings back a finished result: a meeting summary, a deck, a spreadsheet. It works on the desktop, on the web and on the phone, and the session keeps going even with the laptop closed. Included in Pro and above.

One Claude. Since September 16, 2026, Cowork has been folding into the regular chat, and a big task can be handed off from any conversation without picking a mode. The rollout covers Pro and Max; Team and Free get it later. The same day, Claude Docs and Claude Slides showed up in beta on the paid plans: Claude builds a document or a presentation right inside the conversation, and the presentation can be downloaded as PowerPoint.

Claude in Chrome. The extension reads the open page, clicks, and fills out forms; by September 2026 it had left beta and can work without confirmation at every step. Available on paid plans; the vendor itself does not recommend it for banking, passwords, or medical data.

To translate and summarize pages with a choice of models, compare the Sider sidebar. Its card separates page reading from the actions of the Hand agent.

Research and connectors. Research mode on paid plans runs several searches in sequence, each building on the previous one, and compiles a detailed answer from many sources. Connectors to Google Drive, Gmail, Slack, and any MCP server can be connected even on Free, while add-ons for Excel, PowerPoint, and Outlook are available on Pro and above.

What was actually done. In the official videos, the agent compiles a summary, task list, and presentation from a folder of meeting transcripts, while the Chrome extension works through slide comments and edits them in 119 steps. This site is written by the same Anthropic models, but through Claude Code, a coding agent on paid plans; we did not run the chat itself for this task, and the numbers come from official demos and others' tests.

Step 1. Sign up at claude.ai by email or with a Google account. According to user reports, registration may require a phone number from a supported country; +7 is not accepted. The rule is simple: the country of the address, phone number, and card must match.

Apps. iPhone and iPad: App Store, iOS 18 required. Android: Google Play, package com.anthropic.claude. The desktop app for macOS and Windows downloads from claude.com/download, and Cowork installs along with it.

Anthropic's country list and the app stores rechecked Sep 26, 2026 · steps from the vendor's help center and practitioner writeups on vc.ru and Habr, snapshot Sep 5, 2026

Claude is third among the world's chat assistants and growing the fastest.

Market share. According to Similarweb data in Momentic's roundup, in May 2026 Claude had 9.2% of global chat-assistant web visits and 952.6 million visits per month, versus 99.7 million a year earlier, a 9.5x increase. According to First Page Sage, as of September 1, 2026, its US share was 10.2%, versus 51.5% for ChatGPT and 27.6% for Gemini. The methodologies differ, but the picture is the same: third place.

App. On March 2, 2026, Claude Mobile had 11.3 million daily active users versus 250.5 million for ChatGPT (Similarweb, TechCrunch). The gap is more honest than any share: Claude has surpassed Perplexity and Copilot, but is far behind ChatGPT and Gemini.

Models. Opus 5.5 scored 58 points in the Artificial Analysis index and as of Sep 26, 2026 sits in first place. Fable 5.1 and GPT-6 Astra from OpenAI hold 53 each, Opus 5 scored 51. The measurement was made at maximum effort, while the chat defaults to medium. The top-ranked model lives in the chat app that is third by audience.

Its place among the assistants in our catalog is shown under the card's name, and the whole tool ranking sits on a shared page.

Chat is free; you pay for limits and Cowork.

Free entry. Free gives you chat on Sonnet 5.5 and Haiku 4.5 with search, memory, files and five projects; the full set is in the table below. The top models, Cowork and Research stay paid.

Pro. $20 a month, or $17 with annual billing, that is $200 up front. Inside you get five times the limit, Opus 5.5 and Cowork, while Fable 5.1 on Pro comes only for extra credits at API prices.

Max. From $100 a month, with five times the Pro limit. The twenty-times-limit option is listed on the site without a price; in the App Store it sells for $249.99.

Limits. The vendor does not publish message counts: the window is five hours, paid plans also have a weekly cap, and chat and Claude Code draw on one pool. Practitioners estimate Free at 15–40 messages per window and Pro at about 45 short ones, while a single forty-page PDF eats a window in 5–8 turns. These estimates were made before September 22, 2026, when Anthropic raised the five-hour limits on Pro, Max and Team alongside Opus 5.5 and gave subscribers a free limit reset.

Teams. Team starts at $20 per seat per month with annual billing; Enterprise is the same $20 per seat plus usage at API prices. Since February 12, 2026, Enterprise can be bought on the website without calling sales.

Three questions: where your chats go, what gets an account banned, and when the model will refuse.

For comparison, open the Lumo assistant: Proton describes encryption of saved history and model processing of requests separately.

Data and training. Since August 28, 2025 Anthropic trains its models on private users' chats unless you have turned that off in the privacy settings. If you agree, the data is kept for up to five years, a deleted chat is wiped from their systems within 30 days, and incognito chats never go into training. The company is American, so don't upload contracts and personal data here.

Bans. In the second half of 2025 Anthropic blocked 1.45 million accounts worldwide and granted 1,700 of 52,000 appeals, about 3% (Anthropic transparency report, updated Jan 29, 2026).

Refusals and model switching. When asked about exploits or virology, Fable 5.1 and Opus 5.5 hand the conversation to an earlier Opus model. The switching is on by default and can be turned off with a toggle in settings. The bigger precedent: on June 12, 2026, Anthropic pulled Fable 5 from its plans and brought it back on July 1. By August 2026, practitioners call refusals one of the most frequent complaints.

Errors. In a separate article, the vendor acknowledges that Claude invents links and claims it sent an email that did not exist. Verify figures and quotes from its answers.

The choice isn't "who's smarter" but "what you do more often".

Different workflows. ChatGPT remains the generalist: text, code, files, search and images in one window, more than a billion people a week. Claude wins on long documents, careful writing and code, and it does not make images or video. Practitioners who pay for both split it this way: Claude for code and writing, ChatGPT for images and general tasks (Habr, Mar 25, 2026).

Context and models. On paid plans Claude reads up to a million tokens at once versus 256,000 for ChatGPT Plus in thinking mode. Opus 5.5 leads the Artificial Analysis index with 58 points, OpenAI's GPT-6 Astra has 53. In the Opus 5.5 announcement Anthropic compares it with GPT-6 Astra and GPT-5.6 Sol, and Gemini is not there at all.

Pro for Claude and Plus for ChatGPT cost $20, but ChatGPT has the inexpensive Go plan for $8, while Claude has no inexpensive paid plan.

In three years, a chat with access in two countries has grown into an agent's workspace that works in folders and the browser, while models arrive here first with every generation.