Claude.ai (Pro and Max)
10/10The consumer-facing chat product at claude.ai, with Projects (persistent workspaces with uploaded reference files and system instructions), Artifacts (live HTML, React, SVG, Mermaid, .docx, .xlsx, .pptx, .pdf outputs), file-system memory, and remote MCP connectors. Pro at twenty US dollars per month is the entry point; Max at one hundred or two hundred US dollars per month removes the usage ceiling and gives priority model access. Live Artifacts, added in April 2026, refresh against connected data sources so dashboards and trackers update on reopen.
Key features
Projects: persistent workspace with uploaded files, custom instructions, and retrieval over project content rather than naive context dumpingArtifacts: renders code, dashboards, Word, Excel, PowerPoint, and PDF outputs in chat; Live Artifacts refresh against connected dataRemote MCP connectors in Settings (Google Drive, Gmail, Slack, GitHub, Higgsfield, custom URLs) with one-click OAuthFile-system memory written to /memory and read at session start; opt-out in Privacy settingsExtended thinking with xhigh effort level on Opus 4.7 and 4.8 for genuinely hard problems
Best use cases
- Draft a fifteen-page proposal grounded in three previous case studies, an RFP PDF, and a price list, all loaded into one Project
- Build a one-page client dashboard as a Live Artifact that refreshes against a Google Sheet of pipeline data
- Summarise a hundred-page tenancy contract and produce a ranked list of unusual clauses
- Turn a messy Plaud transcript into a structured follow-up email with action owners and dates
- Run a Computer Use task: open Excel, fill in a margin model, save it to a folder
Weaknesses
No native image generation, no native video generation, no voice mode, no native real-time web index of the quality Perplexity or ChatGPT offer. Web search is available but feels like a bolt-on. UK users still get the occasional capacity message on Pro during US working hours.
Pricing
Free tier with daily caps. Pro twenty US dollars per month. Max one hundred US dollars per month (five times Pro usage) or two hundred US dollars per month (twenty times Pro usage). Team Standard twenty-five US dollars per seat per month, Premium one hundred and twenty-five US dollars per seat (includes Claude Code). Enterprise from circa five hundred US dollars per month, custom from there.
Reach for this when the deliverable needs to read like a competent senior wrote it.
Anthropic's agentic coding system, available as a CLI, VS Code extension, JetBrains plugin, macOS and Windows desktop apps, a web surface at claude.ai/code, the Claude Agent SDK (TypeScript and Python), a GitHub Action, and a GitHub App. It reads your repo, makes multi-file edits, runs tests, and iterates on failures. The 2026 step change is dynamic workflows on Opus 4.8: it can spawn hundreds of parallel subagents in one session for codebase-scale migrations.
Key features
Subagents and agent teams: parent session delegates to YAML-defined workers with their own context and toolsSkills: reusable instruction packs the session calls into when triggers match; bundled set plus a 6,700-skill marketplaceHooks: deterministic shell commands the harness runs around tool calls, used for permissions, formatting, automated behaviourPlugins and MCP servers: 9,000-plus extensions across the official marketplace and community sitesDynamic workflows and Ultra Code mode on Opus 4.8 for parallel codebase-scale work
Best use cases
- Migrate a Django 4 codebase to Django 5 across two hundred files in one session
- Stand up a FastAPI service from a written spec, including tests, CI, and a Dockerfile
- Add Stripe billing to an existing Next.js app and write the regression tests
- Run a security review of the current diff as a hook before every git push
- Pair with a non-developer ops manager to script repetitive Excel and Outlook tasks via Python
Weaknesses
Genuine learning curve if you have never lived in a terminal. Token costs add up quickly with parallel subagents; Max plan is effectively required for serious daily use. Plugin fatigue is real, ten heavy MCP servers eat the context window before you have written anything.
Pricing
Included in Pro (limited usage), Max (the realistic plan for daily use at one hundred or two hundred US dollars per month), and Team Premium (one hundred and twenty-five US dollars per seat per month). API usage is metered separately at Opus 4.7 rates (five US dollars input, twenty-five US dollars output per million tokens).
Reach for this when you would otherwise hire a junior developer for two weeks of glue work.
Anthropic's desktop agent for knowledge workers, launched January 2026 and broadened in February 2026 with enterprise connectors (Google Drive, Gmail, Docusign, FactSet, Slack) and domain plugins for finance, engineering, HR, and legal. Cowork has access to a sandboxed shell and to user-selected local folders, so it can organise files, generate documents from local material, schedule recurring tasks, and run multi-step jobs. Computer Use reached general availability on claude.ai in March 2026 and is the engine underneath.
Key features
Local file system access (read, write, edit) within user-approved foldersSandboxed shell for code execution and multi-step chainingRecurring and on-demand task scheduling from the desktop appCustomize panel groups skills, plugins, and connectors in one placeEnterprise connectors for Gmail, Drive, Slack, Docusign, FactSet, and others
Best use cases
- Reorganise five years of project folders against a new naming convention overnight
- Pull the last twenty Docusign contracts, extract renewal dates, and post a summary to Slack
- Schedule a Monday morning task that summarises the previous week's Gmail threads with each named client
- Generate a board pack by pulling the latest figures from FactSet and the deck template from Drive
- Run a research task across five named companies and drop the comparison into Excel
Weaknesses
Newer than Projects and Artifacts, so the rough edges are still there. Computer Use can misclick or take unexpected paths on complex web apps; treat it as a junior assistant that needs a second pair of eyes. Genuinely useful only on Max or Team Premium where the usage budget supports daily agent runs.
Pricing
Included in Pro for limited usage, Max for serious use, and Team Premium for organisations. Enterprise pricing on request.
Reach for this when the work lives across local files, email, and Drive and you do not want to script it yourself.
Claude API (developer platform)
9/10The Anthropic developer API at platform.claude.com. Three live tiers as of May 2026: Opus 4.8 (most capable, agentic, codebase-scale), Sonnet 4.6 (balanced, beats prior-generation Opus on coding evals), Haiku 4.5 (fast and cheap). Features include prompt caching, batch inference, files API, citations, extended thinking, tool use, and the Computer Use beta. The Claude Agent SDK is the recommended path for building agentic apps; the Files API and Batch API matter for any document-heavy workflow.
Key features
Prompt caching at five-times input cost discount on cached tokens, transformative for long-document workflowsTool use and the Claude Agent SDK (TypeScript and Python) for production agent loopsFiles API and citations for grounded document question-answeringComputer Use API for headless agentic browser and OS automationBatch API at fifty percent discount for non-realtime jobs
Best use cases
- Build a back-office document classifier over inbound supplier invoices using Haiku 4.5
- Power a proposal-drafting microservice that grounds answers in a fixed corpus using Sonnet 4.6 with prompt caching
- Run nightly batch summarisation of customer support tickets at half cost
- Embed Claude into a custom CRM where the data cannot leave a UK-hosted environment via AWS Bedrock or Google Vertex
- Build an in-house meeting summariser that takes Plaud transcripts and produces a structured Pydantic output
Weaknesses
Per-token cost is materially higher than Gemini or DeepSeek for equivalent jobs at the Opus tier; you need to be disciplined about routing easy work to Haiku or Sonnet. No native image generation endpoint; you wire in a separate provider. Rate limits on Tier 1 accounts feel tight for any genuine production workload.
Pricing
Haiku 4.5 at one US dollar input, five US dollars output per million tokens. Sonnet 4.6 at three US dollars input, fifteen US dollars output. Opus 4.7 at five US dollars input, twenty-five US dollars output. Opus 4.8 at the same headline pricing as 4.7. Prompt caching, batch, and contextual retrieval each carry separate discounts.
Reach for this when you are wiring Claude into your own product or back-office and want production-grade behaviour.
Claude Mobile and Desktop apps
7/10Native iOS, Android, macOS, and Windows clients. Same model line as the web. The desktop app is where Cowork and Computer Use live; mobile is essentially the web app in a wrapper with voice dictation and photo uploads. Not yet a full voice product in the ChatGPT sense.
Key features
Photo upload from camera for receipts, whiteboards, handwritten notesVoice dictation in (text out, not full voice mode)Push notifications when scheduled Cowork tasks completeSame Projects and Artifacts as the webDesktop is the host surface for Cowork agents and Computer Use
Best use cases
- Snap a photo of a contractor's quote on site and have Claude rebuild it as a clean estimate
- Dictate a rough voice note on the drive home and let Claude turn it into a proper follow-up email
- Pin Claude to the macOS menu bar for one-keystroke access to a Project
- Read a long PDF on the iPad and ask Claude to extract a punch list
Weaknesses
No full duplex voice mode in May 2026, which puts it behind ChatGPT for hands-free use. Mobile experience lags the web for Projects and Artifacts. Push notifications still feel beta.
Pricing
Included in every paid tier; the apps are the same product as the web.
Reach for this when you are on site or in a car and need the same Claude you use at your desk.