GPT-5, Claude, Comet & Cursor: What’s Worth Paying For in 2025
Notes from a degree-educated, 55-year-old Scot who’s spent two decades in e-commerce and tech.
2025 Reality Check
AI isn’t magic; it’s power tools. Use them where they pay.
I’ve seen more “next big things” than I’ve had hot dinners. Some were snake oil. Some—like GPT-5, Claude, the Comet AI Browser, and Cursor—are proper tools. The trick is matching the job to the tool and watching the bill. Here’s how I’d size them up for a busy e-commerce operation running from Black Friday to January returns, without losing the rag or the margin.
ChatGPT-5 (GPT-5): expert-level reasoning, finally practical
What it is: OpenAI’s latest model—stronger reasoning, fewer hallucinations, and better judgement when to “think” versus answer quickly. If you found GPT-4 decent but occasionally woolly, this tightens the screws. Official details: Introducing GPT-5 and GPT-5 for work.
Pros: excellent for drafting product copy, tricky spreadsheet logic, promotion calendars, and complex “what if”s. It’s quicker and more grounded, which is what you want when you’re juggling a Christmas range and a late lorry.
Cons: still not a mind reader; it can slip on simple maths if you don’t pin it down. And some folks miss the chattier vibe from older models—so set your tone rules and stick to them.
Where I’d use it: first drafts of product pages, email/ads variants, customer-service macros, and internal policy/safety wording (with human sign-off). It’s also handy for quick merchandising logic—e.g., “if price < £20 and size = Small, exclude from free-ship banner”.
Claude AI & Claude Code: safe pair of hands, great with long context
What it is: Anthropic’s models with a reputation for safety, coherent long-form writing, and careful reasoning. Their dev-oriented Claude Code mode is excellent for refactoring and explaining unfamiliar code. Official pages: Claude pricing and model token costs.
Pros: better at staying factual in big documents, polite by default, and less likely to wander into daft claims. For policy, T&Cs summaries, returns guidance, or long FAQ rewrites, it’s a belter.
Cons: the free tier is tight; creative flourish can be a bit restrained. Coding mode is strong, but heavy use can get pricey on higher plans or via API tokens.
Cursor: the AI code editor your devs will actually use
What it is: An AI-powered IDE on top of VS Code. It chats about your codebase, proposes diffs, does multi-file edits, and can run “background agents” for longer tasks. Site: cursor.com.
Pros: brilliant for small teams that live in JavaScript/TypeScript/Python. Helps accelerate fixes to Shopify/Liquid snippets, email templates, and warehouse integration scripts without leaving the editor.
Cons: can bog down on giant legacy repos; and pricing recently changed—worth understanding the limits.
Comet (Perplexity) — an AI browser that actually clicks for you
What it is: A new AI-first browser from Perplexity that navigates, clicks and summarises—agentic web work, basically. Page: perplexity.ai/comet.
Pros: fab for market research, sourcing competitor copy, summarising long spec sheets, and gathering pros/cons for supplier shortlists. If your day is fifty tabs deep, Comet aims to be the junior you wish you had.
Cons: currently tied to Perplexity’s top subscription, so not cheap; and as a fresh product, it’ll have rough edges as they scale.
Pros & Cons at a glance
- GPT-5 — Best for tough reasoning and polished drafts; watch maths/detail, set tone rules.
- Claude / Claude Code — Great for long docs & careful code help; can feel cautious; heavier use costs.
- Comet — Automates browsing & research; pricey and new—trial before rollout.
- Cursor — Fast AI help inside your IDE; large repos and pricing nuances.
Costs, briefly (today’s ballparks)
- ChatGPT (GPT-5): free tier with caps; business/enterprise varies by seat—see OpenAI.
- Claude: consumer Pro historically ~£18–$20/m; heavier/enterprise via tokens (see pricing).
- Comet: currently tied to Perplexity Max (~$200/m) — Engadget.
- Cursor: Pro ~$20/m; Ultra ~$200/m — see pricing.
Always check the current page—these things change faster than Scottish summer weather.
What to use when (from someone who’s shipped a few Christmases)
- Marketing & merchandising: Spin up GPT-5 for first-draft copy, promo angles and A/B ad variants. Keep brand tone in a style guide the model must follow. Publish only after a human pass.
- Policy, FAQs, long guides: Use Claude to summarise and restructure. It’s better at staying sensible over long text. Ideal for returns policies and marketplace compliance where one loose claim can cost you.
- Research sprints: Trial Comet for supplier comparisons, competitor audits and spec-sheet summarising. If it saves an hour a day in peak season, the price may wash its face; if not, park it.
- Coding & integrations: Put Cursor on the dev machines. Great for gnarly Liquid templates, Shopify app tweaks, and the “quick fix” that usually eats an afternoon.
Governance, because you’ll thank yourself later
- Prompts & playbooks: save the good ones in a shared doc; version them.
- Publish rules: no AI edits to legal copy/prices without human sign-off. Ever.
- Changelog: log what the model changed and when. Rollbacks save reputations.
- Privacy & children’s data: mind your ICO/EU rules and be transparent when a bot is answering.
It’s not there to spoil the fun; it’s there to stop daft decisions being automated at scale.