Claude

Where Claude sits

Claude is Anthropic's consumer assistant and one of the two products that define the AI assistant category. Anthropic was founded in 2021 by Dario and Daniela Amodei with a group of former OpenAI researchers, and Claude — named for Claude Shannon — arrived in 2023 and has gone through a long model lineage since. The company's differentiating bet has always been safety research and careful behaviour (Constitutional AI and all that), but what a buyer experiences in 2026 is narrower and more practical: Claude is the assistant people reach for when the work involves long documents, careful prose, or codebases, and it has quietly become a developer platform too. Claude Code, the terminal coding agent, is now included on every paid plan, and the current flagship models — Opus, Sonnet and Haiku in their 5.x generations, plus the Fable line — run context windows up to 1M tokens per Anthropic's support documentation.

The product surface has expanded past chat. Claude now ships as web, desktop and mobile apps; a Chrome extension (Claude in Chrome); Microsoft 365 add-ins; Claude Design, Docs and Slides for document work; Research for cited deep-dives; connectors and Skills for bringing in external tools and repeatable procedures; memory across projects; and Claude Cowork for collaborative work. This matters for the review, because "Claude" is no longer just a text box — it is a small suite, and the plan you buy determines which pieces you get.

What it actually does

At its core, Claude writes, analyses, summarises, translates, and codes. In my judgement it is the strongest of the big assistants at sustained long-form work — reading a 200-page document and answering questions about it, drafting and revising a document across many turns without losing the thread, and reasoning through ambiguous instructions with fewer confident inventions than I see elsewhere. The 1M-token context on current models means a stack of PDFs, a large codebase slice, or a full quarter of meeting notes fits in one conversation; Anthropic's docs note that single requests accept up to 600 images or PDF pages. Projects hold instructions and reference files per workflow; Skills store repeatable procedures; connectors link to Google Drive, GitHub, Slack and similar.

Pricing at time of writing, per Anthropic's pricing page and the plan guides I checked: Free is $0 with chat on web, desktop and mobile, connectors, skills and memory, but capped usage, Sonnet/Haiku models only, and a 5-project limit. Pro is $20 a month ($17 a month billed annually, $200 up front) with at least 5x Free usage per 5-hour session, Opus access, unlimited projects, Claude Code, Claude in Chrome, the Microsoft 365 add-ins and Research. Max 5x is $100 a month and Max 20x is $200, monthly only — same models, far more headroom, priority at peak times, and monthly API credits ($100 or $200). Team covers 2-150 people with mixable seat types: Standard at $25 a seat monthly ($20 annually) with 1.25x Pro usage, Premium at $125 ($100 annually) with 5x Standard usage — plus SSO, central billing, shared projects, org-wide skills, usage analytics and a contractual no-training default. Enterprise is $20 per seat per month billed annually plus all usage at API rates, adding SCIM, audit logs, a Compliance API, custom retention, customer-managed keys, IP allowlisting and a HIPAA-ready offering. That Enterprise reshape — from quote-only bundle to published seat-plus-consumption pricing — landed in September 2026.

My testing notes

I have used Claude daily through several model generations, most recently on Pro and Max, across writing, research and coding workloads.

First observation: long-document work is where Claude earns the reputation. In a recent test I dropped a 140-page regulatory document plus three supporting exhibits into a project and asked for a cross-referenced risk summary with section citations. The output mapped claims to the right sections and — the part I watch for — declined to resolve two genuinely ambiguous clauses instead of guessing, flagging them for human review. When I retried the same task with a competitor's assistant at comparable settings, it produced a cleaner-looking summary with two misattributed citations. That refusal-to-invent behaviour has been consistent enough across my Claude use that I now route due-diligence work here first. The 200K window of earlier generations was already useful; the 1M window on current models changes the workflow from "chunk the document" to "load the case."

Second observation: Claude Code is the feature that justifies Max for developers. Running the agent against a real repository — tests to write, a refactor to execute — it planned multi-file changes, ran the test suite, and iterated on failures without hand-holding in ways that saved me hours on routine work. The boundary: on Pro's usage allowance, a couple of heavy sessions can exhaust a 5-hour window; Max 5x removed the interruptions in my use. This is a usage purchase, not a quality purchase — Max runs the same models as Pro — and I would not buy it for someone who chats occasionally.

Third observation: the writing quality is genuinely different in kind, particularly for editing. My standard test is a 300-word product paragraph I want cut to 150 without losing three specific facts. Claude preserved every fact, cut redundancy, and asked whether to keep one clause rather than silently dropping it — a small behaviour that makes it trustworthy as an editor. For code, Claude Code is the star, but plain chat is also strong on review and explanation; the weaknesses are image work (no image generation at all) and anything requiring fresh news or live web data, where Research helps but the baseline chat can be dated.

What it does not do

Claude does not generate images or audio — no DALL-E-style creation, no voice output — and its voice input is limited compared to ChatGPT's. Free-plan usage caps hit quickly and the free tier excludes the tools that matter most (Claude Code, Research, Chrome, the 365 add-ins). Usage limits are the product's constant negotiation: sessions are metered per 5-hour window with weekly caps, and heavy users will bump into them on Pro long before they outgrow the model quality. The Team plan has a two-seat minimum and card-only payment, and mid-year seat adds are prorated while annual prepay refunds do not exist — cancel month three of an annual plan and the remaining months are gone. Enterprise's seat-plus-API-rate model makes bills scale with usage, which is fair but harder to predict; budget caps are essential from day one. And the model roster changes names and tiers frequently (the Fable line's availability differs by plan), so plan-to-feature mapping needs rechecking at renewal.

Verdict

Claude is the assistant I reach for when the work is long, careful, or code-heavy, and the 2026 product line gives individuals and teams real choices without forcing everyone onto the same tier. I rate it 4.4 on the rubric: exceptional depth in analysis, writing and agentic coding, strong value at Pro's $20, a mature and growing platform surface (API, MCP, connectors, Claude Code), held back from the top of the scale by absent image/voice generation, usage ceilings that gate the people who benefit most, and per-seat billing rules that punish casual commitment. If your work involves documents you cannot afford to have misread, it is the most defensible choice in the category; if you need images, voice or unlimited-feeling use on a $20 budget, look elsewhere.

Sources

• https://claude.com/pricing
• https://support.anthropic.com/en/articles/8606394-how-large-is-the-context-window-on-paid-claude-plans
• https://docs.anthropic.com/docs/en/build-with-claude/context-windows
• https://llm-integration.eu/en/claude-pricing-plans-2026
• https://digitavision.com/blog/claude-personal-and-team-plans
• https://agentscamp.com/guides/getting-started/claude-plans-compared-2026
• https://walma.ai/en/guides/ai-tools/claude-plans
• https://justinmckelvey.com/blog/claude-ai-pricing
• https://en.wikipedia.org/wiki/Anthropic
• https://en.wikipedia.org/wiki/Claude_(language_model)